onetaskgraph_github_projects/

lib.rs

//! A stateless onetaskgraph source over one GitHub Projects v2 board.
//!
//! **A write held to a version** (`--if-version`) is checked immediately before its first
//! mutation and not as one step with it: GitHub takes no conditional write, so the engine reads
//! the task through this source just before writing it, and a change landing between that read
//! and the mutation — the time one request takes — is GitHub's residual race.
//!
//! **A board is a container of projects, not a project.** Its own `title`,
//! `shortDescription` and `readme` are never read as an item's fields and are never
//! written: nothing in this source can rename the board a user configured.
//!
//! **A project is an issue and its tasks are that issue's sub-issues.** GitHub's schema
//! decides that: `Issue` exposes `parent`, `subIssues` and `subIssuesSummary`, and
//! `DraftIssue` exposes none of them. Creating an issue needs a `repositoryId`, and a
//! board has none, so a write without [`GitHubProjectsConfig::repository`] is refused
//! naming the field — but that repository is the *fallback*, not the home of every item.
//!
//! <!-- llmlint: ignore-block[contracts_have_one_source_or_a_drift_gate] The rule's one
//! executable source is `GitHubProjectsSource::creation_target`; this is where a reader of
//! the module meets it, and `tests/plugin.rs` drives every arm below against the loopback
//! board and asserts on `createIssue`'s own `repositoryId`, so the prose cannot outlive a
//! change to the rule. -->
//! **Which repository an issue is created in is decided by the item's own `repositories`
//! field, under one rule.** Exactly one entry names the repository the issue is created in:
//! a task issue is where a person finds the work from the repository it changes, and one
//! filed in a board's nominated repository is invisible from every other. Zero entries, or
//! two or more, name none, so a task's or a document's issue is created in the repository
//! its parent project's issue lives in — read from the board, or from this process's own
//! record of a project it created earlier in the same command — and a project's issue, or
//! a task or document written with no parent, is created in the configured `repository:`.
//! What that rule refuses, it refuses before `createIssue`, so no issue is half-created. An
//! existing issue is never moved: the update path leaves the issue where it is and records
//! the list in the metadata slot when it differs, so the read side's derivation and the
//! creation rule agree by construction.
//! <!-- llmlint: ignore-end[contracts_have_one_source_or_a_drift_gate] -->
//!
//! **A document is an ordinary issue whose title begins [`DESIGN_TITLE_PREFIX`].** A
//! board has no document type and nothing but issues to hold one in, so the title is the
//! discriminator and it is the whole of it. The title this source *reports* is the one a
//! person wrote, with the prefix taken off — the same way the metadata slot is taken off
//! the body so `content` is what the person wrote — and writing a document puts the prefix
//! back, so a round trip returns the title that went in.
//!
//! **Telling a document from a project from a task.** The design prefix is read **first**:
//! a document is never a project and never a task, whatever sub-issues it has or does not
//! have. Only then does the rest apply — a board issue is a project when *either* it has
//! sub-issues *or* it carries [`ItemKind::METADATA_KEY`]; otherwise it is a task. A
//! sub-issue is always a task, whatever it carries. The marker is sufficient and never
//! necessary: it is what makes an *empty* project — the state a project copy passes
//! through between creating the project and filing its first task — readable as a
//! project, while the sub-issue arm lets a person author a project on the board by hand
//! with no knowledge of this product's metadata at all. Reading the prefix later than the
//! sub-issue rule would make a design issue with no sub-issues an empty project, which is
//! exactly the state that rule exists to catch. Pull requests are neither a project nor a
//! task nor a document and are ignored.
//!
//! **A task's comments are its issue's comments.** They are read off `Issue.comments` and
//! written with `addComment`, `updateIssueComment` and `deleteIssueComment`, and a comment's
//! id is GitHub's own node id for the `IssueComment`. Two things GitHub decides are refused
//! rather than papered over: a board **draft** is not an issue and has no comments at all, so
//! a comment call on one is refused rather than answered with an empty page; and GitHub signs
//! every comment as the account the token belongs to, so a comment handed an author of its
//! own is refused rather than posted under another name. GitHub's comment mutations take the
//! comment's id and nothing else, so an edit or a delete first reads which issue that comment
//! is on, and a comment on some other issue is one this task does not have.
//!
//! **Where an entity is, is a link.** Every project, task and document this source reports
//! carries a [`Location::Url`] naming the issue's own web address — the same address the
//! `url` field already reports, in the shape that says a reader can open it. That is the
//! contrast the location contract exists for: a reader holding an entity from this source
//! is handed something to link to and one holding an entity from a folder of Markdown is
//! handed a path, and neither has to know which plugin answered. It does not replace or
//! derive from `url`; that field goes on reporting what it always reported.
//!
//! **Where metadata lives.** Short typed things go to typed fields and native relations:
//! status to the board's `Status` single-select and the issue's own state, the copy
//! origin to a source-owned `onetaskgraph.origin` text field, and dependencies to
//! `blockedBy` and to sub-issue links. Unbounded caller JSON goes in a trailing
//! `<!-- onetaskgraph.metadata ... -->` comment at the end of the issue body — the same
//! encoding `docs/metadata.md` settles for Linear, not a second one. A ProjectV2 text
//! field is length-bounded and `shortDescription` is capped at 300 characters, which is
//! why neither can hold a caller's own prose. Setting one caller key on its own — on a task,
//! a project or a document alike — is one update of the issue body that changes that slot
//! and not one byte outside it, and it is not sent at all when the key already holds the
//! value. The link a copy records on an item it copied, `onetaskgraph.copies`, is small and
//! is kept in that same slot, written by that same update.
//!
//! **A caller's value can also be projected onto a board text field**, so the board can filter
//! on it: each [`GitHubProjectsConfig::metadata_fields`] entry names a text field, a top-level
//! metadata key and a path inside its value. The slot stays the value's one home — every read
//! reports metadata from it alone — and the field follows it on every write of an item, a
//! create, an update and a copy alike: a string is written, and not sent again while the field
//! holds it — an empty string included, which is written as itself; nothing there or `null`
//! clears a field holding a value; any other
//! JSON type is refused before any mutation. It rides the ordered [`graphql::UPDATE_FIELDS`]
//! write beside `Status`, `Priority` and the origin, so it adds no request; a key set on its
//! own moves its field first and the body last, as every write of an existing item does. A
//! board without the field, or with a non-text field of that name, is refused before any
//! mutation pointing at `sources fields`, whose [`GitHubProjectsSource::fields`] creates it as
//! a text field and never changes a field that is there.
//!
// llmlint: ignore-block[contracts_have_one_source_or_a_drift_gate] This public module documentation is a required user-facing description; the loopback plugin tests and shared live journey drive StatusMapping resolution, both mutations, and observed read-back together.
//! **Status.** `status_mapping` is per-instance configuration, in the shared grammar
//! [`onetaskgraph_plugin_api::StatusMapping`] documents, from a status category to an
//! option of the board's one `Status` field for a task and for a project: a bare name is
//! the option for both kinds, `null` disables the category for both, and `{task, project}`
//! names it per kind. A category the mapping does not mention keeps its shipped default for
//! both kinds; one it mentions is exactly what it configures, so a per-kind object no longer
//! gets the shipped default for the kind it leaves out. Two categories one kind would read
//! back from one option are refused as the configuration is read, ignoring case, while one
//! option may stand for different categories of the two kinds. Writes go by the kind of the
//! item written: a status that kind has no option for, or whose option the board lacks, is
//! refused before any mutation, naming the source, the kind, the category and the key
//! `status_mapping.<category>.<kind>` — there is no fallback. `done` selects its mapped
//! option and closes the issue as `COMPLETED`; `cancelled` selects its mapped option and
//! closes it as `NOT_PLANNED`, for either kind. Every open category reopens a closed issue
//! before selecting its option. Reads give a closed issue's reason precedence over its
//! option, while an open issue's option decides its category through its own kind's
//! mapping, and an option that mapping does not name reads as `unknown` under its own name.
//! The guarded [`GitHubProjectsSource::status_options`] and
//! [`GitHubProjectsSource::fields`] operations are the one path here that calls
//! `updateProjectV2Field`: GitHub replaces the whole option list, so they preserve every
//! existing option id and verify the field and item assignments immediately afterwards.
//! They ask for both kinds' options, counting a terminal category's mapped option as
//! configured because a terminal write refuses without it. No ordinary source read or
//! write calls that mutation, whose
//! `singleSelectOptions` *overwrites* a field's option set, so no addition is additive
//! and a mistake destroys every item's status. A status this board cannot represent is a
//! refusal naming the status and the instance instead.
//!
//! `unknown` has no shipped option because this source cannot preserve an open-ended
//! status word: it writes an existing board option and never
//! creates an option. An operator may map `unknown` to one existing option, in which case
//! every unknown word lands on that option and reads back as `unknown` under the option's
//! name. This differs from `local-md`, which writes and reads the original word itself.
//!
//! The shipped terminal options are exactly `done: Done` and `cancelled: Cancelled`.
//! `done` also closes the issue because GitHub derives `subIssuesSummary.completed`
//! and the board's own `Sub-issues progress` field from closed sub-issues: a plan whose
//! finished tasks were only moved to a "Done" column would read 0% complete forever.
// llmlint: ignore-end[contracts_have_one_source_or_a_drift_gate]
//!
//! # What this source declares, field by field
//!
//! One verdict per field of [`Capabilities`], and what `Native` means when this source
//! says it. *Proven* means a shared journey drives it against the real
//! binary over this source's own row in `crates/onetaskgraph-e2e-support/src/fixtures.rs`, and
//! `every_row_declares_exactly_what_its_plugin_reports` is what keeps this list and
//! [`capabilities`](TaskSource::capabilities) from parting.
//!
//! | Field | Verdict |
//! | --- | --- |
//! | `projects` | **Supported and proven,** and the one predicate here that is pushed down rather than applied in process: a task's project is the issue it is a sub-issue of, so a listing scoped to one *asks that issue* for its own sub-issues. This is the field that was declared and then not applied, which silently returned another project's tasks. |
//! | `documents` | **Supported and proven.** A board holds issues, so a document is one: the issue whose title begins [`DESIGN_TITLE_PREFIX`]. Reads, filters and paging answer on exactly the terms a task read does, and a write puts the prefix back. |
//! | `comments` | **Supported and proven,** over the task issue's own comment connection, oldest first and paged by GitHub's own cursor; added, edited and removed through GitHub's comment mutations, paced as every other mutation is. A draft item has no comments on GitHub and is refused, and so is an author, because GitHub records the signed-in account as every comment's author. |
//! | `assets` | **Supported and native.** Image references travel as GitHub user attachments in the repository the task or document issue lives in. Uploads are authenticated and verified before the issue body is written. |
//! | `priority` | **Supported and proven** by an instance configured with `priority_mapping`, and declared unsupported by one without it, which reports every task's priority as `none` and sends exactly the requests it sent before priorities existed. The priority is the board's single-select `Priority` field: no value is `none`, a mapped option is its level, matched case-insensitively, and an option the mapping does not name fails the read of that task, naming the option. A write selects the mapped option, or clears the value for `none`; a board without the field or the option is refused, pointing at `sources fields`, which is the one thing that creates either. |
//! | `assignees` | **Supported and proven when `config.assignees: true`,** unsupported otherwise, which reads no issue's assignees and sends exactly the requests it sent before assignees existed. An issue's `assignees` by login, every one it holds, selected on the reads the source already makes; a write replaces them with one login or none as one more aliased field of the document the write already sends, and a login GitHub refuses, or answers leaving off the issue, is refused for that item naming it. See [`assignees`] for the login lookup. Proven by `onetaskgraph-github-projects-e2e`'s `assignees` journeys. |
//! | `filter_by_priority` | **Supported and proven,** over the priority each task reads as — `none` for every task of an instance without `priority_mapping`. |
//! | `filter_by_comment_activity` | **Supported, and exact** for comments created and for comments edited at or after `commented_since`, in every repository — of any owner — the board's items live in. Applied by asking a narrower question rather than by reading the board: GitHub's issue search scoped by `project:<owner>/<number>` alone, with an `updated:>=` qualifier, names the candidates, and each candidate's own comments confirm it, so `ProjectV2.items` is never read. That rests on GitHub moving an issue's `updatedAt` when a comment on it is added **or edited**, which the credentialed journey `an_edited_comment_moves_its_issue_and_is_selected_since` re-takes on every run of this lane. The search is an index that lags a write — the credentialed lane has watched it miss a newly commented issue for thirty seconds — so an issue this process itself commented on in the same command is a candidate whatever the search says, read by its own node if the search did not name it, and has its comments read rather than being ruled out by an `updatedAt` from before the comment. A comment another process wrote is found only once the index has it, so a caller asking again from its last instant should overlap the two generously. |
//! | `orphan_tasks` | **Supported and proven.** A task issue with no `parent` is in no project. |
//! | `filter_by_label` | **Supported and proven,** over the issue's own labels. |
//! | `filter_by_status` | **Supported and proven,** over the board's `Status` option and the issue's open or closed state, through this instance's own `status_mapping` for the item's kind — a task query by the task half, a project query by the project half, `unknown` included. |
//! | `filter_by_metadata` | **Supported, and asked of GitHub** — through the body and never through a projected `metadata_fields` text field, which is the board's to filter on and not this source's to read. A query naming metadata values is one board-scoped issue search with each value a quoted phrase `in:body` — GitHub's index covers the metadata comment at the end of the body, which is where caller metadata lives — and every candidate is confirmed against its own parsed metadata comment, so only an item holding that string at that key and path is returned. **A value with no letter or digit is refused** — the empty string, whitespace or punctuation alone — before any request, as a `SourceError::Refused` (wire kind `refused`) naming the value: GitHub's index holds words, so no bounded query can find such a value, and this source neither reads the whole board for it nor answers it as empty. |
//! | `filter_by_origin` | **Supported, and asked of GitHub without enumerating the board.** The union of three reads, each confirmed by an exact match against the item's own origin field: the board's field filter over the `onetaskgraph.origin` text field, the issue search for the id as a phrase in the body where a write of this release mirrors it, and this process's own writes. See *Where a read-after-write guarantee comes from* for the window the three leave. |
//! | `reconcile_status` | **Supported,** and proven by `onetaskgraph-github-projects-e2e`'s `status_reconcile.rs`: the board's own `ProjectV2.items(query:)` filter, `is:closed`, read page by page and never the unfiltered items, each candidate confirmed from its own item node — a closed task whose card is not at the option the mapping gives its closure's category. |
//! | `complete_by_issue` | **Supported,** and proven by `status_complete.rs` there and in the engine's suite: each issue a reference names read in one aliased request per fifty — its repository, its number, its first page of board memberships with each card's `Status` read by name — and nothing else of the board. |
//! | `search_title` | **Supported, and asked of GitHub for a task,** over `Issue.title`: a task query's text is one board-scoped issue search for it as a phrase `in:title`, every candidate confirmed by the case-insensitive substring rule. GitHub matches whole words, so a task holding the text only inside a longer word is not returned — a narrowing this source declares rather than hides. **A text with no letter or digit that is not blank is refused** — `--` for one — before any request, as the same `refused` error naming the text, for the reason a metadata value like it is; a blank text is not refused, and keeps the board read it always had, confirmed by the same substring rule. A project query's text, and a document query's text when the query is scoped to no project, is that same board-scoped search for the same phrase in the same fields, refused on the same terms, every candidate confirmed by its kind and by the same substring rule, so it narrows exactly as a task's does; a document query scoped to one project sends no search, reads that project's sub-issues and confirms its text over them by the substring rule alone, so it is neither narrowed to whole words nor refused for a text with no letter or digit. A board draft is not an issue, so no text search lists one, a draft titled as a document included. |
//! | `search_content` | **Supported,** on the same terms, `in:body`, over the visible body — the trailing metadata comment is not part of what the substring rule confirms. |
//! | `task_dependencies` | **Supported and proven,** in both directions: `blockedBy` and `blocking`. |
//! | `project_dependencies` | **Supported and proven,** in both directions, over the same two connections, because a project here is an issue. |
//! | `max_page_size` | **Supported and proven.** [`MAX_PAGE_SIZE`], GitHub's own connection maximum. |
//!
//! Image writes use `POST https://uploads.github.com/user-attachments/assets` with `name`,
//! `content_type` and the issue repository's numeric `repository_id` as query parameters,
//! the source token as `Authorization: Bearer`, the image content type as `Content-Type`,
//! and the raw bytes as the request body. The numeric repository id is read at most once
//! per repository per source instance. A loopback API endpoint also selects the loopback
//! uploads host; production needs no assets repository, commits or extra configuration.
//! Every returned JSON `url` is read with the same token before a body is written: only
//! a 2xx verification succeeds. A refusal names the asset, URL and HTTP status; an
//! anonymous 404 does not invalidate an authenticated 200. An upload refusal names the
//! asset and HTTP status and says the token type may not be accepted by the upload
//! endpoint. A classic PAT was measured accepted; no acceptance claim is made about
//! fine-grained PATs or OAuth tokens.
//! The body keeps the authored alt text while `./<name>` becomes the attachment URL,
//! and `onetaskgraph.assets` records `{sha256, url}` by name. A matching destination
//! SHA-256 reuses its URL without an upload or verifying read; changed bytes are uploaded
//! and verified afresh, in the repository the issue already lives in on an update.
//! An uploaded image renders for the viewers GitHub lets read it, like the issue text beside it.
//! Uploads take the same mutation spacing as content writes. Both uploads and verification
//! reads wait out classified rate limits within the configured per-call retry budget,
//! on the supplied clock; verifying reads take no mutation slot.
//!
//! Nothing here is unsupported. `documents` and `comments` are not predicates — they say this
//! source has documents and that its tasks have comments, both of which hold — and the three
//! facts behind the uniform `Native` on the
//! predicates beside it are recorded below rather than re-derived, because a reader who
//! takes `Native` to mean *the remote service filters* will read that uniformity as a
//! lie.
//!
//! First, the plugin contract defines `Support::Native` as *the source applies this
//! predicate itself*, and says nothing about where it applies it. What the declaration
//! promises the engine is capability rule 1 — a predicate declared `Native` **is** applied
//! — so that the engine may push it down and apply nothing of its own.
//!
//! Second, this source can keep that promise for every predicate at no additional API
//! cost, because whichever of the reads below answers a query has already read every
//! candidate that query will return before it filters anything. Filtering those items is
//! in-process work over data already in hand.
//!
//! Third, six task predicates are asked of GitHub as a narrower question and the rest are
//! applied in process over what that question returned. A project filter has a relationship — a
//! project's tasks are that issue's sub-issues, and asking the issue for them is both cheaper
//! and exact. Comment activity is the issue search's `updated:` qualifier. A text search, and
//! a search for metadata values, is the board-scoped issue search carrying the text and each
//! value as quoted phrases; an origin is the board's own field filter over its origin field
//! beside the same search for the id. **The text search narrows, and that is this source's
//! declared semantics:** GitHub matches whole words where the substring rule this source and
//! the local Markdown source confirm with would match inside one, so an item holding the text
//! only inside a longer word is never a candidate. Every item returned does contain the text.
//! A project query's text, and a document query's scoped to no project, is that same search
//! and narrows on the same terms, its candidates confirmed by their kind as well.
//! GitHub's issue search offers no qualifier for a label set, a status column or a priority,
//! so those three are applied in process over the candidates, and a query carrying none of
//! the six narrowing predicates reads the board. Declaring one `Unsupported` would make the
//! engine compensate for work this source has already done, and declaring `projects` native
//! while ignoring the filter (which this source once did) silently returns another project's
//! tasks, because the engine trusts the declaration and applies nothing locally.
//!
//! # The three ways this source reaches an item, and what each costs
//!
//! A board read is charged for what its *nested* connections could return rather than for
//! what was asked, so one whole-board read costs the same whether the question was about
//! one project or about all of them. That is why a question about one project is never
//! answered by reading the board:
//!
//! | The question | What is sent | What it costs |
//! | --- | --- | --- |
//! | one item, by its own id | [`graphql::ISSUE`] — `node(id:)`, carrying the field definitions of the boards it sits on and the far ends of its `blockedBy`, which is what a write of it needs — and, when that node is a board draft, [`graphql::DRAFT`] — the draft and the one board item it is | the item |
//! | one task with its first page of comments, for `task show` and a comment listing | [`graphql::ISSUE_DETAIL`] — the same `node(id:)` read with the issue's `comments` | the item and a page of its comments |
//! | one board page with each task's first comments, for `TaskSource::query_task_details` | [`graphql::BOARD_WITH_COMMENTS`] — the board's own `items`, without search completion | tasks and first comment pages in one request; board consistency, with comment-activity filtering refused |
//! | several tasks with their comments, for `task show-many` | [`graphql::ISSUE_DETAILS`] — [`DETAIL_BATCH`] aliased `node(id:)` fields per request | each item and a page of its comments |
//! | the board's own id and field definitions, for a write whose item does not carry them | [`graphql::BOARD_FIELDS`] — the board's `id` and `fields`, and no `items` — or, for a create that needs the repository's id too, [`graphql::CREATION_CONTEXT`], both in one request | the board's fields |
//! | one project's tasks or documents | [`graphql::SUB_ISSUES`] — that issue's own `subIssues` | that project |
//! | which projects this board holds | [`graphql::SEARCH_ISSUES`] — an issue search scoped to the board | the board's issues, without their board items |
//! | which projects hold a text, or which documents do when no project narrows the question | [`graphql::SEARCH_ISSUES`] — the board-scoped search with the text as one quoted phrase, `in:title`, `in:body` or both, as a task's text is sent — walked to its end in pages of twenty | the issues that match |
//! | which tasks were commented on since an instant | [`graphql::SEARCH_ISSUES`] — the same board-scoped search with an `updated:>=` qualifier — then [`graphql::ISSUE_COMMENTS`] for each candidate it names | the issues updated since, and their comments |
//! | which tasks hold a text, or a metadata value | [`graphql::SEARCH_ISSUES`] — the board-scoped search with the text and each value as quoted phrases, `in:title`, `in:body` or both, and an `updated:>=` qualifier too when comment activity is asked for — in pages of twenty, only as many as the caller's rows need | the issues that match |
//! | which tasks were copied from one origin | [`graphql::ORIGIN_LOOKUP`] — the board's own `items` under its field filter on the origin field, and the same board-scoped search for the id `in:body`, in one request, each paged at three | the carriers of that origin, which is one item |
//! | every task, every document, every label, when nothing above narrows the question | [`graphql::BOARD`] — the board's own `items` — **and** [`graphql::SEARCH_ISSUES`], because neither enumeration of a board is complete alone; see [`GitHubProjectsSource::board`] | the board, twice over |
//! | which board item one issue is, past the page that came with it | [`graphql::ISSUE_BOARD_ITEMS`] — that issue's own `projectItems` | one issue's memberships |
//!
//! The following standalone-ticket requests are pinned by the real CLI fixture journeys
//! `follow_up_writes_resolve_each_item_once_and_batch_the_copy_fields` and
//! `a_batched_detail_read_costs_one_request_and_one_point_per_detail_batch`, as request count
//! equal to declared points equal to the row. They include the origin lookup and the
//! field/repository discovery a create needs. A bound re-copy changes status, priority,
//! content and metadata; comment recount means a subsequent detail read. Each request here
//! costs one declared point. A membership beyond the embedded page can additionally require
//! the one-point membership recovery described above. A bound re-copy of a task filed under a
//! project adds one read, the engine confirming that project's link by its own id once per
//! command; and the same-source far ends a write newly names — those that do not already block
//! the item, whose own read answered for them — are read together by their own ids,
//! [`DETAIL_BATCH`] to one [`graphql::ISSUE_DETAILS`] request, each new edge then one
//! [`graphql::ADD_BLOCKED_BY`]. Both additions are rows of the table below, pinned by
//! `a_bound_recopy_adds_one_project_read_and_batches_the_dependencies_it_newly_names`.
//!
//! **[`DETAIL_BATCH`] is 24**: the largest batch of [`graphql::ISSUE_DETAILS`] the node-count
//! model prices at one point. Each aliased item is six of GitHub's aggregate, so 24 are 144,
//! which rounds to one point, and 25 are 150, which rounds to two; `tests/point_cost.rs`
//! holds both halves.
//!
//! **An existing item is written body last.** A bound re-copy and a `task update` send its
//! board fields first — the `Status` option and the `Priority` together, in one request — then
//! its parent and its `blockedBy`, and its title, body and state in one `updateIssue` last.
//! GitHub runs no two requests as one, and runs a document's mutation fields in order without
//! undoing an earlier field when a later one fails, so that order is what makes a write
//! refused part-way leave the item's body, and every metadata key in it, exactly as it stood;
//! the one piece of metadata written before the body, an origin a copy re-points, is put back
//! when a later write is refused — and when putting it back is refused too, the write's own
//! refusal names that key, what it now holds and what it held. `crates/onetaskgraph-github-projects-e2e/tests/e2e/write_order.rs` refuses each
//! of those writes in turn, whole and as one aliased field failing after the one before it.
//!
//! **Two facts about GitHub the write rows rest on, each read off GitHub's published schema
//! artifact <https://docs.github.com/public/fpt/schema.docs.graphql> on 2026-10-01 and pinned
//! in `tests/fixtures/schema.graphql`, and the first then put to GitHub itself:**
//!
//! - **A board is accepted at creation but its item is not answered, so a create still files
//!   the issue itself: a new copy is 5 requests, and 4 with `--create`.**
//!   `CreateIssueInput.projectV2Ids: [ID!]` is declared there — "An array of Node IDs for
//!   Projects V2 associated with this issue", `@possibleTypes(concreteTypes: ["ProjectV2"])`.
//!   The credentialed journey `real_projects_v2_contract_writes_and_leaves_no_residue` was run
//!   against a real board on 2026-10-01 with a create sending the board there and reading the
//!   item off the payload's `Issue.projectItems`: every one of its four creates answered with
//!   no item on the board, so each went on to [`graphql::ADD_TO_BOARD`], and the fourth was
//!   refused "Content already exists in this project" — GitHub had filed the issue after
//!   answering, and refuses a second filing rather than answering with the item it holds. A
//!   create therefore sends no `projectV2Ids` and files the issue with
//!   `addProjectV2ItemById`, the one call whose answer names the board item. The saving that is
//!   real is the read before it: the board's fields and the repository's id together, in
//!   [`graphql::CREATION_CONTEXT`], at the point the repository is known.
//! - **A comment still reads its target first, so a comment is 2 requests.**
//!   `AddCommentInput.subjectId: ID!` is declared there with
//!   `@possibleTypes(concreteTypes: ["Issue", "PullRequest"], abstractType:
//!   "IssueOrPullRequest")`. A board draft is no such subject and would be refused, but a
//!   project's issue, a document's issue, an issue on no board of this source and a pull
//!   request all are: GitHub writes the comment, so there is no refusal to map into "that is
//!   not a task of this board". [`graphql::ISSUE`] before [`graphql::ADD_COMMENT`] is what
//!   refuses those by name.
//!
//! | Verb | Requests / points | Documents |
//! | --- | --- | --- |
//! | new copy | 4 | ORIGIN_LOOKUPS (the origin, the board's fields and the repository's node and numeric ids together, a page of one carrier for each origin), CREATE_ISSUE filing the issue on the board and under its project, CREATED_BOARD_ITEMS once GitHub has added it, WRITE_FIELD_VALUES |
//! | copy --create | 4 | CREATION_CONTEXT, CREATE_ISSUE, CREATED_BOARD_ITEMS, WRITE_FIELD_VALUES: the new copy with no origin to ask about |
//! | bound copy | 3 | ISSUE (with the board's fields and the issue's `blockedBy`, so no BOARD_FIELDS or ISSUE_DEPENDENCIES), UPDATE_FIELDS when a board field differs, then UPDATE_ISSUE last carrying only the title, body or state that differs — none, when nothing does |
//! | bound copy, filed under a project | 4 | the bound copy's three, and one ISSUE of the destination project its link names, read once per command |
//! | bound copy, newly naming n dependencies | + ceil(n / DETAIL_BATCH) + n | ISSUE_DETAILS for the far ends that do not already block the item, DETAIL_BATCH (24) to a request (one alone is ISSUE), then one ADD_BLOCKED_BY each; a far end already blocking it is answered by its own read and costs nothing |
//! | comment | 2 | ISSUE, ADD_COMMENT: the target is read first, because GitHub accepts a comment on any issue or pull request (see below) |
//! | detail | 1 | ISSUE_DETAIL: the item and its first page of comments, for `task show` and `task comment list`; `--no-comments` is ISSUE alone |
//! | batched detail | ceil(n / DETAIL_BATCH) | ISSUE_DETAILS: `task show-many` of `n` items, DETAIL_BATCH (24) at a time, comments included or not |
//! | copy --comments, n changed tasks | ceil(n / COPY_DETAIL_BATCH) | ISSUE_COPY_DETAILS: each task with its first page of comments and its `blockedBy`, COPY_DETAIL_BATCH (21) at a time; a task whose comments run past that page adds one ISSUE_DETAIL per further page, and a task skipped from its listing costs nothing beyond it |
//! | recount | 1 | ISSUE_DETAIL |
//! | status | 2 | ISSUE, UPDATE_FIELD; a terminal status additionally updates issue state |
//! | priority | 2 | ISSUE, UPDATE_FIELD or CLEAR_FIELD, with stored priority in the mutation response |
//! | content | 2 | ISSUE, UPDATE_ISSUE |
//! | metadata | 2 | ISSUE, UPDATE_ISSUE |
//! | document metadata, `--expect` a version | 2 | ISSUE, then UPDATE_ISSUE only when the issue is still at that version — the precondition GitHub's own update cannot take — primed with what the read answered |
//! | update | 3 | `task update` naming any of title, body, metadata, status and priority — all five included: ISSUE, UPDATE_FIELDS (the status option and the priority together), UPDATE_ISSUE (title, body with its metadata slot, and state) last |
//! | record only | 1 | ISSUE |
//!
//! <!-- github-search-paging:start -->
//! Board-scoped text, metadata, project-name and comment-activity searches send every
//! page at `first = 20` (SEARCH_PAGE_SIZE), the SEARCH_ISSUES document's one-point
//! ceiling. A later page is sent only when `hasNextPage` is true and the caller still
//! needs rows. A page is never resized to the rows still needed: GitHub orders one
//! search differently at different page sizes, so one fixed size makes a paged walk
//! send exactly the requests one whole read sends, and the answer's order is the order
//! those pages arrive in. A page below twenty would cost the same one point, and GitHub
//! prices this document by rows, so twenty-row pages cost per row what 100-row pages do.
//! Project-name lookup continues until an exact match or exhaustion. A task limit bounds
//! returned and fetched pages: a limit is sliced from the pages it needs, and local
//! confirmation can require more candidates than matching rows. Walking all pages
//! returns the whole answer. The opaque version-4 source cursor carries GitHub's page
//! cursor and how far into that page the last answer stopped, and resumes in the same
//! process or a new one, without duplicates or gaps. It carries no rows: one process
//! sends each page's search once, and a new process re-reads only the page it resumes
//! in, then sends a further page once, never as a re-read, only when its limit still
//! needs rows. Every request either walk sends is the one a whole read sends for that page. Own writes replace stale index
//! copies and complete missing rows at exhaustion. Cache entries are whole GitHub pages,
//! so a small answer cannot truncate a wider question. Origin pages remain three; whole-board sizing is unchanged.
//! Read-after-write is a per-process guarantee. A cursor resumed in a new process is
//! not required to include the original process's writes still omitted by the index.
//! <!-- github-search-paging:end -->
//!
//! The board half of an issue — its board item's id, its `Status` option and this
//! source's origin text field — rides along on `Issue.projectItems` in the first three, so
//! an item reached any of those ways resolves through the same
//! [`GitHubProjectsSource::resolve`] the board walk uses and reports the same title, the
//! same status, the same labels and the same qualified id. That connection comes back a
//! *page* at a time, at `BOARD_ITEMS_PAGE_SIZE`, so the entry for this board is looked for
//! on the page in hand and — only if that page reports more of the connection — in the
//! last row's read of that one issue's memberships, resumed from the page's own cursor and
//! walked to exhaustion. An issue with no entry for *this* board is not this source's to
//! report, which is what keeps an id naming another repository's issue from being answered
//! as an item of this board; and because the page is where the search starts rather than
//! where it ends, that answer is one about a connection read to exhaustion and never about
//! an unread page. Nothing costs the extra read but an issue on more boards than a page
//! holds: an issue this board really does not hold reports no next page, so its
//! memberships are already exhausted where they arrived.
//!
//! **No document here selects the board's own `Labels` field, and nothing is lost by
//! that.** An item's labels are read from its content alone, wherever that content is
//! reached: the three documents above select `Issue.labels` on the fragment, and
//! [`graphql::BOARD`] selects the same connection on the `... on Issue` arm of its
//! `content`. A board's `Labels` field is not one anybody fills in: it is a built-in
//! `ProjectV2FieldType`, it is absent from `ProjectV2CustomFieldType` so no project can
//! create one, and `ProjectV2FieldValue` — the whole of what
//! `updateProjectV2ItemFieldValue` accepts — offers no way to write one. So GitHub derives
//! it from the content, for every content type it exists on, and there is nothing it can
//! hold that the content does not already say: for an `Issue` it *is* that issue's own
//! labels, so selecting it beside them unions a set with itself.
//!
//! **A draft loses nothing by that either**, which is the reasoning this paragraph once had
//! backwards. `DraftIssue` exposes no `labels` field, and by the three schema facts above
//! it cannot carry a board `Labels` value to be derived from one — so a draft has nothing
//! to select *and nothing to lose*, and reports no labels at all. A `PullRequest` item is
//! discarded by [`GitHubProjectsSource::resolve`] before labels are read. Both halves are
//! held to that by tests in `tests/plugin.rs`: the four ways an item is reached report one
//! label set, and that set is the fixture issue's own, by
//! `an_item_reports_the_same_labels_title_status_and_id_however_it_is_reached`; and a board
//! item whose content is a draft reports an empty set, by
//! `a_board_item_whose_content_is_a_draft_reports_no_labels_at_all`. The absence of the
//! selection is held over [`graphql::DOCUMENTS`] by
//! `no_document_selects_the_boards_own_labels_field`.
//!
//! The whole-board row is still the board's own item connection, and deliberately: a
//! **draft** board item is not an issue, so no search can list one, and the reads that have
//! to answer for the whole board are the ones whose cost is the board's size anyway.
//!
//! **A question about one item this source already names by id never lists the board.**
//! Whether that item is on this board, and what its board fields are, is answered by reading
//! that item — its own `Issue.projectItems`, walked to exhaustion by
//! [`GitHubProjectsSource::resolve_issue`], or a draft's own board item — and never by
//! looking for it in [`graphql::BOARD`]'s `items` or in a listing this command already
//! holds. That covers a write's destination, the project a new item is filed under, a
//! same-source far end a dependency names, a status write, the dependency slot a draft keeps,
//! and the delete that takes back an item a copy made. What such a write needs of the board
//! and the item does not carry — the board's id, the `Status` and origin field definitions —
//! comes from [`graphql::BOARD_FIELDS`], which reads no item at all. The reason is evidence,
//! not economy alone: `ProjectV2.items` is a projection that lags the membership GitHub
//! itself reports — an issue added with `addProjectV2ItemById` can be missing from it for
//! minutes. Scanning this host's 842-item board has refused a document copy and an update
//! even though the items' own reads named that board. A scan there gives the wrong answer
//! as well as paying for every page. So a `board.items` lookup does not belong on any of
//! those paths.
//!
//! **What a read may return is capped too, and that cap is on the document rather than on
//! the board.** GitHub limits the number of nodes **one query may return** to
//! [`NODE_COUNT_LIMIT`] and refuses a query above that before executing it: the answer is
//! an error naming the connection the count crossed at, not a slow or a partial result.
//! Every board this source reads is refused the same way, so no board is too big for these
//! documents and none is small enough to save one that is over.
//!
//! The count is arithmetic over the document's own text: each connection contributes the
//! `first:` it asks for, counts **multiply** down a nested path and **sum** across sibling
//! paths. Those are [GitHub's published rules][node-limits] and this workspace does not
//! restate them — `github-graphql-node-count` implements them, and
//! [`worst_case_node_count`] under [`largest_page_sizes`] is where every node count here
//! comes from. `every_document_this_source_sends_stays_under_githubs_node_limit`, in
//! `tests/node_count.rs`, recomputes every document in [`graphql::DOCUMENTS`] from that
//! same text on every run and fails naming any that reaches the limit — so a connection
//! added to a shared fragment is caught there rather than by GitHub.
//!
//! What decides those counts is the page sizes: [`MAX_PAGE_SIZE`] on the outer page,
//! `NESTED_PAGE_SIZE` on the connections hanging off one item, and
//! `BOARD_ITEMS_PAGE_SIZE` on the page of an issue's board memberships a read carries.
//! `$nestedFirst` is spent twice down one path of a board read, so that constant is
//! effectively squared there, which is why it is the one the limit is most sensitive to.
//! `BOARD_ITEMS_PAGE_SIZE` is small for a reason of its own, recorded beside it: what a
//! page of memberships misses is recovered by one further read rather than refused, so it
//! buys a bound every read pays for at the price of a request only a multi-board issue
//! pays.
//!
//! **`nodeCount` and `cost` are two numbers against two limits, and both are computed
//! offline here — per document, one document at a time.** `nodeCount` is the one above: the
//! most nodes one query may return, checked per query and bounded by [`NODE_COUNT_LIMIT`].
//! `cost` is rate-limit points, metered per hour across everything one credential does; it
//! is what the two limiters [`Limiter`] tells apart meter, and a document under
//! [`NODE_COUNT_LIMIT`] still says nothing about its price. [`worst_case_point_cost`] is
//! that second number, and `tests/point_cost.rs` pins every document in
//! [`graphql::DOCUMENTS`] at what it costs — there being no per-call point ceiling to hold
//! one under, the pin itself is the check. The credentialed lane reconciles both figures
//! against GitHub's own, off a probe it already sends.
//!
//! **What is pinned that way is a per-document price and never a session's.** The record in
//! `session-cost.md` measures the two quantities a whole session can be counted in offline —
//! **requests** and **worst-case nodes** — and neither is points. What one whole session
//! consumes of the hourly point allowance is observable only from a credentialed run's own
//! `x-ratelimit-*` headers, which is what [`accounting`] fills its per-budget figures from
//! and what `tests/live.rs` prints at the end of every run.
//!
//! [node-limits]: https://docs.github.com/en/graphql/overview/rate-limits-and-node-limits-for-the-graphql-api
//!
//! **Where a read-after-write guarantee comes from, since neither of GitHub's two
//! enumerations of a board can supply one alone.** Resolving a node id is strongly
//! consistent, so a read by id and a project's own sub-issues are already current. The
//! other two are not, and they are behind by different amounts and in different directions:
//!
//! - GitHub's **issue search** is an index and answers a write made moments ago with the
//!   value from before it — usually for a second or two.
//! - **`ProjectV2.items`** is a projection GitHub rebuilds behind the write, and an item put
//!   on a board with `addProjectV2ItemById` can be **absent** from it — not present with its
//!   content withheld, absent, with the connection walked to its own `hasNextPage: false` —
//!   for *minutes*, while `Issue.projectItems` names the same membership at once.
//!
//! That second one is a measurement rather than a caution. This repository's own
//! credentialed journey writes a project and waits for the board to report it, then writes a
//! task and waits for the same thing seconds later on the same board: the project wait is
//! answered through the search and converged in two or three attempts in each of three runs,
//! and the task wait is answered through `ProjectV2.items` and converged in none of them
//! inside thirty. Separately, an item added to a second and larger board was read back by
//! `Issue.projectItems` on that board's own id while every one of that connection's nine
//! pages, walked to exhaustion nine minutes after the add, did not name it. Reading a board
//! through the lagging one alone is what had a board read deny an issue that had certainly
//! landed on it.
//!
//! So [`GitHubProjectsSource::board`] is the **union** of both — each search result still
//! admitted only on this board's own strongly-consistent `Issue.projectItems`, and neither
//! enumeration dropped, because only `ProjectV2.items` lists a board draft and only the
//! search reports what the projection is behind on. What closes the last
//! gap, the one where both are behind, is [`GitHubProjectsSource::created`]: every read this
//! source answers is completed with what this process itself wrote, so an item created
//! seconds ago is reported whether or not GitHub has caught up. Nothing else is remembered,
//! nothing is written down, and the record dies with the process. **A wait that has to
//! observe GitHub's own data cannot be answered from that record** — which is why the
//! credentialed journey asks through a source built afresh, and why the union above rather
//! than a longer wait is what makes such a wait converge.
//!
//! **A narrowed read is the same bargain, stated for each of the three predicates it
//! answers.** A read carrying a text, metadata or origin predicate asks GitHub's index rather
//! than walking the board, and every such answer is completed with what this process wrote —
//! its [`created`](GitHubProjectsSource::created) record and every existing item it wrote,
//! each filtered by the same predicates as the rest — so an item this command wrote a moment
//! ago is returned by a query that matches it whether or not the index has caught up. An item
//! a caller holds the id of is read by that id, with `node(id:)`, which is strongly
//! consistent. What is left is stated rather than papered over:
//!
//! | Read | Finds | Behind by |
//! | --- | --- | --- |
//! | text, metadata | the issue search for the phrases | what another process wrote in the last second or two, until GitHub indexes it |
//! | origin, first read | the board's field filter over the origin field — every carrier, whichever release wrote it | what `ProjectV2.items` is behind on, which the measurements above put in minutes |
//! | origin, second read | the issue search for the id in the body, where a write of this release mirrors it | a second or two, as any search |
//! | origin, third read | this process's own writes | nothing |
//!
//! So an origin carrier another process added within the last second or two, before either
//! index has it, can be missing from an origin query, and one written by the release before
//! this one — its origin in the field alone — can be missing for as long as the board's own
//! item connection is behind on it. A copy that must not duplicate its own earlier write
//! relies on the link it records, not on either index. **A board draft is not an issue**, so
//! a draft is never returned by a text, metadata or origin query, whatever it holds: no search
//! lists one, the origin lookup drops any the board's own field filter names, and one this
//! process wrote is not added back either.
//!
//! **The origin lives in the board field, and the body holds a mirror of it.** A write that
//! carries an origin writes it to the `onetaskgraph.origin` text field and also into the
//! body's metadata slot, so the issue search can find it in seconds. The field is
//! authoritative: this source reads an item's origin from the field alone, so a slot that
//! disagrees with it, or holds one where the field holds none, is never read as a second
//! origin — and the release before this one reads the slot, drops that key's copy for the
//! field's, and sees the same one origin.
//!
//! Filtering happens before paging, so a page of a filtered result is a page of the
//! survivors rather than the survivors of a page. Label matching and the substring rule a
//! text candidate is confirmed by answer the same question the same way the local Markdown
//! source's do; which candidates a text search has to confirm is GitHub's word match, which
//! is the one place the two sources can answer the same text differently.
//!
//! <!-- llmlint: ignore[contracts_have_one_source_or_a_drift_gate] The declaration itself
//! has one source, `capabilities`, and the note above is the reasoning behind it rather
//! than a second copy of it: without the three facts recorded here a reader takes the
//! uniform `Native` for a lie and reverts it. The drift gate on the declaration is this
//! crate's own capabilities test, which pins every field of it against a fully spelled-out
//! `Capabilities` literal — a struct with no `Default`, so a field added to the contract
//! fails to compile there rather than going unasserted. -->
//! The fixture-server tests above run wherever this crate is selected; the credentialed
//! lane runs in the same required check, beside them, and can fail it — it verifies the
//! current schema, then drives every field of the table above against the real board. It builds its own fixture there — two projects, one task filed under each,
//! one filed under neither, a label on one of the three and a closed status on another —
//! because that shape is what tells an honoured predicate from an ignored one: a board
//! holding a single project answers a project filter the same way whether or not this
//! source applies it, which is exactly how the defect above went unseen.
//!
//! That lane writes only to the board `GH_PROJECTS_OWNER` and `GH_PROJECTS_NUMBER` name,
//! and the scratch repository `GH_PROJECTS_REPOSITORY` names:
//! `nickderobertis/onetaskgraph-live-scratch`. It refuses the core repository before any
//! session or request, independently of the live demand. Fix that variable in the machine's
//! onetaskgraph `secrets.env` or the environment it pushes from, such as ai-orchestrator's
//! `.env`. The targets are declared once in `onetaskgraph_github_live`, the lane's own
//! policy crate; absent credentials or nominations otherwise skip.
//!
//! GitHub Actions artifacts carry `ci-<run id>-<attempt>-<micros>`, naming their writing
//! run and attempt; invalid Actions identity refuses before writing. Other runs retain
//! `<host>-<process>-<micros>`. Own cleanup matches the whole stamp. Machine residue stays
//! the owning machine's lock sweep's; Linear always keeps that form and is unaffected.
//! The hourly janitor is cleanup, never a test lane: scratch CI residue is removed only
//! after its run reads back as `completed`, immediately before the listed-artifact batch.
//! Failed or incomplete ownership reads preserve residue. A 24-hour waiting period
//! is a margin, never the ownership authorisation. Cleanup leaves machine stamps
//! and the board's `onetaskgraph.origin` field untouched.
//!
//! # What a session of requests costs, and where the report is
//!
//! This source records **every** request it sends into [`accounting::Accounting`], at
//! `send_once` — the one place a request leaves this crate, which is why a read path added
//! later is counted without anybody remembering to count it. That is the whole of what this
//! crate adds to the arrangement; [`accounting`] is where what a record carries, how a
//! session's spend is arrived at, and what it deliberately does not know are set out.
//!
//! What one whole session of the live journey costs, counted that way against this crate's
//! loopback fixture board, is written down in `session-cost.md` beside this crate — with the
//! reduction it came out of, and with what it does and does not say about rate-limit points.
//!
//! [`GitHubProjectsSource::accounting`] is the read: a snapshot to hold and compare, which
//! [`accounting::Session::report`] renders the session report from. It is on the ordinary
//! code path — no environment variable, no feature, no build configuration — because an
//! instrument nobody switches on measures nothing, and
//! [`Plugin::build_recording_into`] is how a caller making its own calls beside this
//! source's counts the whole session rather than this source's share. The credentialed lane
//! in `tests/live.rs` does exactly that, and prints the report at the end of every run,
//! passed or failed.
//!
//! **A live session refuses to start unless the account can afford it.** Before it does any
//! of the work it exists to do, the journey makes one request — `GET /rate_limit`, which
//! GitHub documents as not counting against the REST rate limit and which answers both of
//! its budgets at once — and starts only if, for each of them, what remains minus this
//! session's estimated cost is still at least
//! `onetaskgraph_live::RETAINED_BUFFER` — twenty per cent — of that budget's whole
//! allowance. A session that cannot **declines**: it did not run, so it is
//! neither a pass nor a failing assertion, and it says which budget was short, that budget's
//! limit, what remained, the estimate, the buffer and when it resets — then stops, without
//! waiting for the budget to come back. The estimate is derived offline from
//! `tests/fixtures/session-cost.txt` and a cost model stated in `tests/journey/budget.rs`,
//! which is also where the published rule that model rests on is cited; the accounting
//! above records the gate's own read like any other request, and
//! [`accounting::Session::report`] prints the estimate beside what the session really spent.
//!
//! **GitHub is the authority on both of its own numbers, and the credentialed lane goes and
//! asks it.** Everything above computes `nodeCount` and `cost` offline from a document's own
//! text, which is what lets it run on every platform and on a pull request from a fork with
//! no credential — and that is what actually stops a regression merging. But an offline
//! arithmetic can only ever agree with itself: if GitHub changes its rules, this workspace
//! goes on computing the old answer and nothing notices. So `tests/live.rs` reconciles them.
//! GitHub's schema exposes `rateLimit(dryRun: true)`, whose `nodeCount` is *"the maximum
//! number of nodes this query may return"* and whose `cost` is what that document would
//! spend, both for a document **without executing it**, and the lane asks it for every query
//! document this source sends, under the largest bindings this source sends, and fails when
//! GitHub's figure and [`worst_case_node_count`] or [`worst_case_point_cost`] disagree. A
//! mutation is skipped, because `rateLimit` is a field of `Query` and cannot be asked about
//! one; the offline pins still cover it. It records what those calls reported about the
//! account's own allowance, because whether asking is free is a thing to observe rather than
//! to assume. Two quantities, not one: [`NODE_COUNT_LIMIT`] bounds `nodeCount` per query,
//! and `cost` is metered against an hourly allowance the accounting above reads off a
//! credentialed run's own response headers.
//!
//! **GitHub has two rate limiters and this source is refused by both, so nothing here
//! treats them as one thing.** The primary budget is the hourly allowance `gh api
//! rate_limit` reports; the secondary limiter is a burst limiter over content-generating
//! requests, and *nothing* reports it. Which one refused decides the operator's next step,
//! so [`Limiter`] is a type rather than a detail, and it is what [`MIN_MUTATION_INTERVAL_MS`],
//! [`GitHubProjectsSource::board_cache`] and [`GitHubProjectsSource::graphql`] each answer
//! one part of.
#![deny(missing_docs)]

use std::collections::BTreeMap;
use std::sync::{Arc, Mutex};
use std::time::Duration;

use chrono::{DateTime, Utc};
use onetaskgraph_plugin_api::{
    Capabilities, Classification, ClosedAs, Comment, CommentBody, CommentOrigin, Completion,
    ContradictedItem, Coverage, Cursor, DependencyEdge, DependencyEndpoint, DependencyKind,
    DependencySupport, Designator, Direction, Document, DocumentQuery, DocumentVersionCoverage,
    Health, HeldTask, IssueHolding, IssueRef, ItemKind, ItemWrite, Label, LabelFilter, Location,
    MetadataKey, Metering, NativeId, NewComment, Page, PageRequest, Priority, Project,
    ProjectFilter, ProjectQuery, ProjectVersionCoverage, Repository, SecretResolver, SharedClock,
    SourceError, SourceName, SourcePlugin, Status, StatusCategory, StatusContradiction,
    StatusMapping, Support, Task, TaskDetailRead, TaskQuery, TaskRef, TaskSource, TaskUpdate,
    TaskUpdateOutcome, TextFields, TextQuery, UnmappedStatus, UpdatedField, VersionCoverage,
    WaitCause, WriteJournal, WriteSupport, system_clock,
};
use reqwest::{Client, StatusCode, Url};
use schemars::{Schema, schema_for};
use secrecy::{ExposeSecret, SecretString};
use serde::{Deserialize, Serialize};
use serde_json::{Value, json};

pub mod accounting;
mod assets;
mod assignees;
mod batch;
pub mod change_token;
mod expected;
pub use expected::{CONTEXT_VERSION, WHOLE_MEMBERS, context_shapes};
mod pacing;
mod projection;
mod relations;
mod visibility;

use accounting::Accounting;

/// The registry name for this plugin.
pub const KIND: &str = "github-projects";
/// GitHub's maximum connection page size.
pub const MAX_PAGE_SIZE: u32 = 100;
/// Every page of a board-scoped narrowing search: 20 rows, one point of SEARCH_ISSUES, the
/// most one point buys. GitHub prices that document by rows, so pages of 20 cost what pages
/// of 100 cost per row, and a page of fewer than 20 costs the same one point.
pub const SEARCH_PAGE_SIZE: u32 = 20;
/// How many items one [`graphql::ISSUE_DETAILS`] request reads, each with the first page of
/// its comments: the largest batch the node-count model prices at one point.
///
/// Each aliased item is resolved once, and what GitHub charges for it is the connections
/// under it — its labels, its page of board memberships, the field values of each of those
/// three memberships, and its comments: six requests' worth of the aggregate GitHub divides
/// by a hundred and rounds. Twenty-four items come to 144, which rounds to one point;
/// twenty-five come to 150, which rounds to two. `tests/point_cost.rs` prices the document at
/// one point and fails if one item more would still be priced at one.
pub const DETAIL_BATCH: usize = 24;
/// How many issue references one [`graphql::ISSUES_BY_REFERENCE`] request reads.
///
/// Each reference asks GitHub for one connection — a page of its board memberships, each with
/// its `Status` value read by name rather than through a connection of field values — so fifty
/// come to fifty requests' worth of the aggregate GitHub divides by a hundred, which is its
/// one-point minimum, and the batch a landing's closures are read in costs one point per fifty
/// issues per source. `tests/point_cost.rs` prices the document at one point.
pub const REFERENCE_BATCH: usize = 50;
/// How many items one [`graphql::ISSUE_COPY_DETAILS`] request reads: each with the first page
/// of its comments and its `blockedBy`, the largest batch the node-count model prices at one
/// point.
///
/// The six requests' worth an item of [`DETAIL_BATCH`] costs, and its `blockedBy` connection
/// besides: seven. Twenty-one items come to 147, which rounds to one point; twenty-two come to
/// 154, which rounds to two. `tests/point_cost.rs` prices the document at one point and fails
/// if one item more would still be priced at one.
pub const COPY_DETAIL_BATCH: usize = 21;
/// How many mutations one aliased request of a copy's creates, closes and updates carries:
/// [`graphql::CREATE_ISSUES`], [`graphql::CLOSE_ISSUES`] and [`graphql::UPDATE_ISSUES`].
///
/// Eleven is what GitHub was seen to accept in one request — the timing spike
/// (`board-copy-speed-2026-10-18-spike-copy-timing-report`) sent eleven aliased mutations of four
/// kinds, three `createIssue`, three `addProjectV2ItemById`, three field updates, an
/// `addSubIssue` and an `addBlockedBy`, and every one ran — and no more was established.
pub const MUTATION_BATCH: usize = 11;

/// The most people one lookup ([`graphql::user_ids`]) asks for: GitHub's documented limit of
/// [`NODE_COUNT_LIMIT`] nodes a request, each `user(login:)` field being one. Every person a
/// call names that this process has not resolved is looked up in one request below it — so a
/// call is split only past it.
#[expect(
    clippy::cast_possible_truncation,
    reason = "500,000 fits every target's usize"
)]
pub const PEOPLE_PER_LOOKUP: usize = NODE_COUNT_LIMIT as usize;

/// The most relation writes — `addSubIssue`, `removeSubIssue`, `addRelatesTo`,
/// `removeRelatesTo` — one request carries: a 60-alias `addRelatesTo` answered HTTP 504 after
/// landing 58 of its 60 (`spike-relations`), and 30 were accepted every time.
pub const RELATION_BATCH: usize = 30;
/// How many aliased writes one request of a copy's board field values and blocked-by edges
/// carries: [`graphql::WRITE_FIELD_VALUES`], [`graphql::ADD_BLOCKED_BYS`] and, for a request
/// holding both kinds, [`graphql::WRITE_FIELDS_AND_EDGES`].
///
/// Nineteen is the most aliases the feasibility spike
/// (`board-copy-gaps-2026-10-24-spike-feasibility-report`) sent in one mutation request and
/// saw GitHub price at the one point every mutation request costs; no larger request was
/// established, so a copy holding more is split.
pub const WRITE_BATCH: usize = 19;
/// How many issues one [`graphql::DELETE_ISSUES`] request deletes: one request of eleven
/// aliased `deleteIssue` met `RESOURCE_LIMITS_EXCEEDED` from the eighth alias on, and batches
/// of five succeeded (the feasibility spike's undo-batching run).
pub const DELETE_BATCH: usize = 5;
/// How many origins one [`graphql::ORIGIN_LOOKUPS`] request asks about: the nine records of a
/// plan and the project's document, and no more, since each carries a page of the board's field
/// filter and one of the issue search.
pub const ORIGIN_BATCH: usize = 10;
/// How many searches one [`graphql::SEARCH_BATCH`] request asks: what `task find` asks of a
/// board in one call, each search an aliased `search` field of its own.
pub const SEARCH_BATCH: usize = 20;
/// How many origins one [`graphql::SEARCH_BATCH`] request looks up beside its searches, each by
/// the board's own field filter in an aliased `items` field of its own and by one of the
/// request's searches.
pub const SEARCH_BATCH_ORIGINS: usize = 5;
/// The rows each search of a [`graphql::SEARCH_BATCH`] request asks for on its first page.
///
/// A batched find asks narrowed questions that each expect a handful of answers, and every
/// search slot is priced at its page size: three rows a slot keep a first round of
/// [`SEARCH_BATCH`] searches at three points, where a page of [`SEARCH_PAGE_SIZE`] each would
/// cost twenty. A search with more to answer is walked on at [`SEARCH_PAGE_SIZE`] rows a page.
pub const SEARCH_BATCH_FIRST_PAGE: u32 = 3;

/// The rows each resumed page of one [`graphql::SEARCH_BATCH`] round asks for, when `resumed`
/// of its slots — searches and origin item listings alike — resume from a cursor.
///
/// Every resumed page of a round shares one [`SEARCH_PAGE_SIZE`] of rows, rather than each
/// asking that many: a round in which all twenty searches walked on at twenty rows a slot cost
/// 22 points in one request, while sharing holds every round to what its first pages cost and
/// one page more. A search with more to answer is walked on over more rounds, each page still
/// read, so what a walk answers is unchanged; a round of one resumed search asks the page a
/// search alone asks.
#[must_use]
pub fn search_batch_resumed_page(resumed: usize) -> u32 {
    let resumed = u32::try_from(resumed.max(1)).unwrap_or(u32::MAX);
    (SEARCH_PAGE_SIZE / resumed).max(1)
}

/// The page sizes of the most expensive [`graphql::SEARCH_BATCH`] round this source sends:
/// every slot on its first page, at [`SEARCH_BATCH_FIRST_PAGE`] rows, but one resumed slot
/// asking all of [`search_batch_resumed_page`]'s shared rows.
///
/// `tests/point_cost.rs` holds every round shape [`search_batch_resumed_page`] allows to cost
/// no more than this, and the credentialed lane reconciles GitHub's own price of the document
/// under it. [`largest_page_sizes`] still binds every slot at [`SEARCH_PAGE_SIZE`], an upper
/// bound on this, so the offline worst case covers more than any round sends.
#[must_use]
pub fn search_batch_largest_round() -> Variables {
    let mut sizes = largest_page_sizes();
    for slot in 0..SEARCH_BATCH {
        sizes.insert(format!("first{slot}"), SEARCH_BATCH_FIRST_PAGE);
    }
    for slot in 0..SEARCH_BATCH_ORIGINS {
        sizes.insert(format!("itemsFirst{slot}"), SEARCH_BATCH_FIRST_PAGE);
    }
    sizes.insert("first0".to_owned(), search_batch_resumed_page(1));
    sizes
}
/// How many comments one [`graphql::ADD_COMMENTS`] request adds, and how many issue updates one
/// [`graphql::WRITE_TASKS`] request carries: [`MUTATION_BATCH`], the most aliased mutations
/// GitHub was seen to accept in one request.
pub const COMMENT_BATCH: usize = MUTATION_BATCH;
/// How many board field values one [`graphql::WRITE_TASKS`] request carries, beside
/// [`MUTATION_BATCH`] field clears: [`WRITE_BATCH`].
pub const TASK_FIELD_BATCH: usize = WRITE_BATCH;

/// The most nodes any one document this source sends may be asked to return.
///
/// GitHub's own published per-query ceiling, taken from
/// [`github_graphql_node_count::NODE_LIMIT`] rather than written out again here, so this
/// workspace cannot hold a stale copy of somebody else's number. A query above it is
/// **refused before it is executed**, whoever is asking and whatever board they are
/// asking about — so this is a bound on the documents rather than a budget that runs out.
///
/// This is `nodeCount`, the maximum number of nodes *one query may return*. It is not
/// `cost`, the rate-limit points a call spends against an hourly allowance shared by
/// everything the credential does — two numbers against two limits, and this constant
/// bounds only the first. The second is computed offline too, per document:
/// [`worst_case_point_cost`], pinned for every document in [`graphql::DOCUMENTS`] by
/// `tests/point_cost.rs`, and reconciled against GitHub's own `cost` by the credentialed
/// lane. There is no constant like this one to hold a price under, because points are an
/// hourly allowance rather than a per-call bound.
///
/// Neither is a session's price. What `session-cost.md` records of a whole session is its
/// **requests** and its **worst-case nodes**; what a whole session spends in points is
/// reported only by a credentialed run's own `x-ratelimit-*` headers, through
/// [`accounting`]. The module section on the three ways this source reaches an item says how
/// the count is arrived at, and which of the page sizes below decide it.
pub const NODE_COUNT_LIMIT: u64 = github_graphql_node_count::NODE_LIMIT;

/// Nested connection size for the connections that hang off one item.
///
/// It multiplies through every document that reaches an item under a page — the count
/// rules multiply down a nested path — so it is the constant [`NODE_COUNT_LIMIT`] is most
/// sensitive to. `tests/node_count.rs` is what holds the pair together: it recomputes
/// every document under these constants and fails naming any that reaches the limit, so
/// raising this is caught there rather than by GitHub.
const NESTED_PAGE_SIZE: u32 = 50;
/// How many of one issue's board memberships are read when an issue is reached directly.
///
/// An issue reached through a search or through its own node id carries its board half in
/// `Issue.projectItems`, and only the entry for *this* board is read. This connection sits
/// under a page of issues, so every point of it multiplies through the whole document and
/// is paid for whether or not any issue is on a second board — which is why it is
/// deliberately far smaller than [`NESTED_PAGE_SIZE`].
///
/// **Three, because what a page misses is now recovered rather than refused**, and the
/// recovery is what the value is chosen against. An issue whose entry for this board sits
/// past this page costs one further request — [`graphql::ISSUE_BOARD_ITEMS`], resumed from
/// that page's own cursor — so the value trades a bound every read pays for a request only
/// a multi-board issue pays. At one, a deployment whose issues commonly sit on two or more
/// boards would pay that request *per issue*, which is order N against the one page per
/// hundred issues a read costs today. At three it is only reached by an issue on four or
/// more boards at once, which keeps the recovery path exceptional rather than routine for
/// a plausible deployment.
const BOARD_ITEMS_PAGE_SIZE: u32 = 3;
/// How many carriers of one copy origin one page of [`graphql::ORIGIN_LOOKUP`] asks each of
/// its two connections for.
///
/// An origin names one item, so the answer an origin lookup expects is one carrier, and a
/// second is a duplicate a copy already takes the first of. Both connections are walked to
/// exhaustion whatever this is, so it decides how many requests an unusual answer costs and
/// never what the answer is. It is small because every point of it is paid on every lookup,
/// and a copy makes one lookup per item it has no link for: at three, ten lookups cost fewer
/// worst-case nodes than the one whole-board read they replaced.
const ORIGIN_PAGE_SIZE: u32 = 3;
/// How many carriers of each origin the batched [`graphql::ORIGIN_LOOKUPS`] asks each of its two
/// connections for: one, the one carrier an origin names.
///
/// A second carrier still shows: the page then says there is more, and that origin is left to
/// its own [`graphql::ORIGIN_LOOKUP`], which walks every page at `ORIGIN_PAGE_SIZE` — so a
/// duplicate is found exactly as before, by the lookup that reads all of them. The feasibility
/// spike (`board-copy-gaps-2026-10-24-spike-feasibility-report`) priced the batch of ten
/// origins at two points with a first page of three and at one with a first page of one.
const ORIGIN_LOOKUPS_PAGE_SIZE: u32 = 1;

/// How many of one project's sub-issues one [`graphql::SUB_ISSUES`] page asks for: the most
/// one point buys, since every row carries six connections GitHub charges for once per row.
/// `session-cost.md` derives it; `tests/point_cost.rs` prices the document at one point and
/// fails if one row more would still be priced at one.
pub const MEMBER_PAGE_SIZE: u32 = 23;

pub use github_graphql_node_count::{NodeCountError, Variables};

/// The largest value this source can bind to each page-size variable its documents name.
///
/// Every `first:` in [`graphql`] reads one of these seven, and each is capped at the
/// constant above it wherever a caller's own limit could reach it — `$first` at
/// [`MAX_PAGE_SIZE`], `$commentFirst` at [`MAX_PAGE_SIZE`], `$nestedFirst` at `NESTED_PAGE_SIZE`, `$boardItems` at
/// `BOARD_ITEMS_PAGE_SIZE`, `$originFirst` at `ORIGIN_PAGE_SIZE`, `$lookupFirst` at
/// `ORIGIN_LOOKUPS_PAGE_SIZE`, `$memberFirst` at [`MEMBER_PAGE_SIZE`]. So this is the worst case a caller can drive this source to,
/// not one configuration of it, which is what makes a bound computed under it a bound on
/// every read.
pub fn largest_page_sizes() -> Variables {
    let mut sizes = Variables::from([
        ("first".to_owned(), MAX_PAGE_SIZE),
        ("commentFirst".to_owned(), MAX_PAGE_SIZE),
        ("nestedFirst".to_owned(), NESTED_PAGE_SIZE),
        ("boardItems".to_owned(), BOARD_ITEMS_PAGE_SIZE),
        ("originFirst".to_owned(), ORIGIN_PAGE_SIZE),
        ("lookupFirst".to_owned(), ORIGIN_LOOKUPS_PAGE_SIZE),
        ("memberFirst".to_owned(), MEMBER_PAGE_SIZE),
    ]);
    // Each slot of the batched search pages at most what one search does.
    for slot in 0..SEARCH_BATCH {
        sizes.insert(format!("first{slot}"), SEARCH_PAGE_SIZE);
    }
    for slot in 0..SEARCH_BATCH_ORIGINS {
        sizes.insert(format!("itemsFirst{slot}"), SEARCH_PAGE_SIZE);
    }
    sizes
}

/// The most nodes `document` could be asked to return, by GitHub's published rules.
///
/// Computed offline from the document's own text under [`largest_page_sizes`] — no
/// network, no credential and no schema — by
/// [`github_graphql_node_count::node_count`], which is where the rules themselves live.
/// A document at or above [`NODE_COUNT_LIMIT`] is one GitHub refuses before executing, so
/// this is what a check holds every document in [`graphql::DOCUMENTS`] below.
///
/// # Errors
///
/// Returns the calculation's own [`NodeCountError`] when `document` does not parse, holds
/// no single operation, or binds a page size this source does not name — each of which is
/// a defect in the document rather than a number.
pub fn worst_case_node_count(document: &str) -> Result<u64, NodeCountError> {
    node_count(document, &largest_page_sizes())
}

/// The most rate-limit points one call of `document` could spend, by GitHub's published
/// rules.
///
/// Computed offline from the document's own text under [`largest_page_sizes`] — no
/// network, no credential and no schema — by
/// [`github_graphql_node_count::point_cost`], which is where the rules themselves live.
/// This is `cost`, metered **per hour** against the allowance one credential shares across
/// everything it does; it is not `nodeCount`, which is [`worst_case_node_count`] and is
/// bounded per query by [`NODE_COUNT_LIMIT`]. There is no per-call ceiling to hold this
/// under, so what `tests/point_cost.rs` does with it is pin every document in
/// [`graphql::DOCUMENTS`] at what it costs, and the credentialed lane reconciles those
/// figures against GitHub's own reported `cost`.
///
/// # Errors
///
/// Returns the calculation's own [`NodeCountError`] when `document` does not parse, holds
/// no single operation, or binds a page size this source does not name — each of which is
/// a defect in the document rather than a number.
pub fn worst_case_point_cost(document: &str) -> Result<u64, NodeCountError> {
    github_graphql_node_count::point_cost(document, &largest_page_sizes())
}

/// The rate-limit points one call of `document` spends under `variables`, by GitHub's published
/// rules: [`worst_case_point_cost`] is this under [`largest_page_sizes`], and a request this
/// source sent is priced by it under the page sizes that request really bound — what GitHub
/// charges the request, since it charges every connection at the page size asked for.
///
/// # Errors
///
/// Returns the calculation's own [`NodeCountError`] when `document` does not parse, holds
/// no single operation, or binds a page size `variables` does not name.
pub fn point_cost(document: &str, variables: &Variables) -> Result<u64, NodeCountError> {
    github_graphql_node_count::point_cost(document, variables)
}

/// The most nodes `document` could be asked to return under `variables`.
///
/// [`worst_case_node_count`] is this under [`largest_page_sizes`], and the accounting in
/// [`accounting`] is this under the bindings one request really sent — one spelling of the
/// calculation, so a bound checked offline and a cost recorded at run time cannot come to
/// disagree. The rules themselves live in [`github_graphql_node_count::node_count`].
///
/// # Errors
///
/// Returns the calculation's own [`NodeCountError`] when `document` does not parse, holds
/// no single operation, or binds a page size `variables` does not name.
pub fn node_count(document: &str, variables: &Variables) -> Result<u64, NodeCountError> {
    github_graphql_node_count::node_count(document, variables)
}

/// `variables` with the page sizes and flags a token twin's inline issue selection binds,
/// at a read's page sizes.
fn token_variables(mut variables: Value) -> Value {
    if let Some(object) = variables.as_object_mut() {
        object.insert("nestedFirst".to_owned(), json!(NESTED_PAGE_SIZE));
        object.insert("boardItems".to_owned(), json!(BOARD_ITEMS_PAGE_SIZE));
        object.insert("duplicates".to_owned(), json!(true));
    }
    variables
}

/// The input members a mutation names the issue or board item it writes by.
const WRITTEN_NAMES: [&str; 4] = ["id", "issueId", "subjectId", "itemId"];

tokio::task_local! {
    /// The source a targeted update is running on, by its address, while it runs: what has its
    /// mutations sent as the twins selecting the whole issue, so the update answers the version
    /// its own last response shows (see `GitHubProjectsSource::written_tokens`).
    static SPELLING: usize;
}

/// The issue-title prefix that makes a board issue a document.
///
/// A GitHub Projects board has no document type — it holds issues — so the discriminator
/// is the title, and this is the whole of it: an issue whose title begins with these bytes
/// is a document and every other issue is the task or project the sub-issue rule makes it.
///
/// It is spelled **once**, here, and read rather than restated everywhere else — including
/// by the shared journeys, which take it from this constant so a board fixture cannot
/// drift from what this source reads. `docs/metadata.md` records the two consequences that
/// are not obvious from the bytes: the reported title has this prefix taken off, exactly
/// as the body's metadata slot is taken off `content`, and this prefix is read *before*
/// the sub-issue rule, so a design issue with no sub-issues is never an empty project.
pub const DESIGN_TITLE_PREFIX: &str = "DESIGN: ";

/// The most characters GitHub stores in one issue's body: a longer one is refused as
/// `body is too long (maximum is 65536 characters)`.
pub const BODY_LIMIT: u64 = 65_536;

/// The address a new upload's URL is spelled at, as GitHub serves a user attachment: what the
/// size of a body is measured with before an asset of it is uploaded, a UUID long.
const PREDICTED_ATTACHMENT_URL: &str =
    "https://github.com/user-attachments/assets/00000000-0000-0000-0000-000000000000";

/// Exact GraphQL query documents issued by this plugin.
///
/// Keeping the production documents here lets the pinned-schema test validate the same
/// bytes that are sent to GitHub, rather than a test-only copy which could drift
/// independently. [`STATUS_OPTIONS_UPDATE`] is the sole document that may rewrite a board
/// field, and its guarded caller always supplies the complete existing option set with ids.
pub mod graphql {
    /// The board half of one item: the field values every document here reads it from.
    ///
    /// A macro for the same reason [`board_issue!`] below is one, a level further in. This
    /// selection is needed by that fragment, by [`BOARD`] under the board's own `items`,
    /// and by [`ISSUE_BOARD_ITEMS`] under a membership walk — and all three have to produce
    /// *the same value*, because
    /// [`GitHubProjectsSource::resolve`](super::GitHubProjectsSource) reads them through
    /// one path. Three spellings of it is what would drift, so there is one.
    ///
    /// The `Status` option and this source's own origin text field are the whole of it. It
    /// selects no `ProjectV2ItemFieldLabelValue`: GitHub derives that field from the item's
    /// content, so it holds nothing the content's own `labels` do not already say, and it
    /// would sit a label connection two page sizes deep.
    macro_rules! board_item_values {
        () => {
            r#"fieldValues(first:$nestedFirst){nodes{
          ... on ProjectV2ItemFieldSingleSelectValue{name field{
            ... on ProjectV2SingleSelectField{id name options{id name}}
          }}
          ... on ProjectV2ItemFieldTextValue{text field{... on ProjectV2Field{id name}}}
        }pageInfo{hasNextPage}}"#
        };
    }

    /// The variable every document reading an issue declares for the relationship events its
    /// change token counts, with its value as the default, so no caller binds it.
    ///
    /// A variable rather than a literal list because every argument a production document
    /// passes is a declared variable, which `tests/schema.rs` holds them to, and every member of
    /// its default to the pinned enum. Its default is the six event types GitHub records a
    /// relationship added or removed with: a dependency on either end, as `blockedBy` and
    /// `blocking`, so one a person makes from the far end moves this end's count too; and a
    /// project membership on the task, as its parent, which GitHub records there whichever
    /// issue the move was made from. A sub-issue's events on its parent are not counted: a
    /// project carries no task token, and a copy writes nothing of a task its own sub-issues
    /// decide. [`super::change_token`] says why the count is in the token.
    macro_rules! relationship_events_variable {
        () => {
            ",$relationshipEvents:[IssueTimelineItemsItemType!]=[BLOCKED_BY_ADDED_EVENT,BLOCKED_BY_REMOVED_EVENT,BLOCKING_ADDED_EVENT,BLOCKING_REMOVED_EVENT,PARENT_ISSUE_ADDED_EVENT,PARENT_ISSUE_REMOVED_EVENT]"
        };
    }

    /// Everything this source reads about one issue, wherever it reaches that issue.
    ///
    /// A macro rather than a constant so the three documents below can `concat!` it: one
    /// spelling of these fields is what makes an issue read through the board-scoped
    /// search, through its own node id, and through its project's sub-issue relationship
    /// resolve to *the same* item, which is the whole of what
    /// [`GitHubProjectsSource::resolve_issue`](super::GitHubProjectsSource) relies on.
    ///
    /// `projectItems` is what carries the board half of an issue: the board item's own id
    /// and the [`board_item_values!`] above — the `Status` option and this source's origin
    /// text field — that a `ProjectV2.items` read used to carry. It is asked for on the
    /// issue rather than on the board, which is what makes the cost of a read proportional
    /// to what was asked for instead of to the board's size.
    ///
    /// It carries a *page* of that connection, at `BOARD_ITEMS_PAGE_SIZE`, and its
    /// `endCursor` is what [`ISSUE_BOARD_ITEMS`] resumes from when this board's entry is
    /// not on that page: a page here is where the search for the entry starts rather than
    /// where it ends.
    ///
    /// It does **not** select the board's `Labels` field value, and that is the whole of
    /// what keeps the three documents below under [`NODE_COUNT_LIMIT`](super::NODE_COUNT_LIMIT):
    /// a label connection there sits under `fieldValues` under `projectItems` under a page
    /// of issues, spending `$nestedFirst` twice down one path, and took
    /// [`SEARCH_ISSUES`] and [`SUB_ISSUES`] to 2,556,100 nodes against a limit of 500,000.
    /// No label is lost — this is a fragment `on Issue`, whose own `labels` are selected
    /// above, and that connection is where every label this source reports comes from. No
    /// document in this module selects the board field any longer, [`BOARD`] included; the
    /// module documentation records why nothing it could have held is lost.
    macro_rules! board_issue {
        () => {
            concat!(
                r#" fragment BoardIssue on Issue{"#,
                board_issue_fields!(),
                "}"
            )
        };
    }

    /// What [`board_issue!`] selects of an issue, written out where a document selects it
    /// inline rather than through the fragment.
    macro_rules! board_issue_fields {
        () => {
            concat!(
                r#"__typename id number title body url createdAt updatedAt state stateReason(enableDuplicate:$duplicates) repository{nameWithOwner} parent{id} subIssuesSummary{total} commentCount:comments{totalCount} relationshipEvents:timelineItems(itemTypes:$relationshipEvents){totalCount}
      labels(first:$nestedFirst){nodes{id name color}pageInfo{hasNextPage}}
      projectItems(first:$boardItems){nodes{id updatedAt project{id number}
        "#,
                board_item_values!(),
                r#"}pageInfo{hasNextPage endCursor}}"#
            )
        };
    }

    /// One page of a repository's issues, newest first, each with when it was created and its
    /// body: what a retried copy reads to find the issue its earlier attempt created, by the
    /// create key in that body's metadata slot — never through search, whose index lags.
    pub const ISSUES_BY_CREATION: &str = "query($owner:String!,$name:String!,$first:Int!,$after:String){repository(owner:$owner,name:$name){issues(first:$first,after:$after,orderBy:{field:CREATED_AT,direction:DESC}){nodes{id createdAt body}pageInfo{hasNextPage endCursor}}}}";
    /// Every issue of one board, found by a search scoped to that board.
    ///
    /// This is how the projects a board holds are listed, and it selects no `items`
    /// connection on `ProjectV2`: the board is a *qualifier of the search* rather than a
    /// container walked page by page, so nothing nested inside a board item is paid for.
    /// Which of the issues it returns is a project is then read off `parent` — GitHub
    /// accepts `-has:parent` as a search qualifier and silently ignores it, so the
    /// discriminator has to be applied to the field, which is a scalar on the issue and
    /// costs nothing.
    pub const SEARCH_ISSUES: &str = concat!(
        r#"query($search:String!,$type:SearchType!,$first:Int!,$after:String,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){
      search(query:$search,type:$type,first:$first,after:$after){
        pageInfo{hasNextPage endCursor}
        nodes{__typename ...BoardIssue}
      }
    }"#,
        board_issue!()
    );

    /// What a dependency read selects of each far end: enough to say which kind of item it
    /// is, its body included for the kind marker.
    macro_rules! related_issue {
        () => {
            " fragment Related on Issue{id title body parent{id} subIssuesSummary{total}}"
        };
    }

    /// One issue by its own node id, which is what a qualified id names here — with what a
    /// write of it needs and the issue does not carry in `board_issue!`: the field
    /// definitions of the boards it sits on, and the far ends of its `blockedBy`.
    ///
    /// Strongly consistent, unlike the search above: GitHub's issue search is an index and
    /// answers a write made moments ago with the value from before it, and resolving a node
    /// id does not.
    ///
    /// **Why those two ride here and not on the fragment.** A copy or an update of an item
    /// reads it by its own id, and with them that one read answers everything the write
    /// needs: which option ids the board's `Status` and `Priority` fields hold — so no
    /// [`BOARD_FIELDS`] — and which issues block it, with each one's kind — so no
    /// [`ISSUE_DEPENDENCIES`]. On `board_issue!` they would sit under the hundred-issue
    /// pages of [`SEARCH_ISSUES`] and [`SUB_ISSUES`], multiplying both documents' price. Here
    /// they sit under one item, and this read is still one point.
    pub const ISSUE: &str = concat!(
        r#"query($id:ID!,$first:Int!,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){
      node(id:$id){__typename ...BoardIssue ... on Issue{
        boards:projectItems(first:$boardItems){nodes{project{id number fields(first:$nestedFirst){nodes{
          ... on ProjectV2SingleSelectField{__typename id name options{id name}}
          ... on ProjectV2Field{__typename id name dataType}
        }pageInfo{hasNextPage}}}}}
        blockedBy(first:$first){nodes{...Related}pageInfo{hasNextPage endCursor}}
      }}
    }"#,
        board_issue!(),
        related_issue!()
    );

    /// One document by its own node id, as [`ISSUE`] reads one, with the project it is filed
    /// under and that project's members beside it: the project issue whole, and a page of its
    /// sub-issues, each as [`SUB_ISSUES`] carries one but with no page of what blocks it.
    ///
    /// What a guarded write of a document's metadata reads, and what reading a document by its
    /// id is: a write to a board the boundary screens is held to who may read the document,
    /// which is the project's and every member's classification as much as its own, and here
    /// the one request that holds the version the write is compared against answers all of it.
    /// Its members carry no edges, which nothing classifying them reads.
    pub const DOCUMENT_IN_PROJECT: &str = concat!(
        r#"query($id:ID!,$first:Int!,$memberFirst:Int!,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){
      node(id:$id){__typename ...BoardIssue ... on Issue{
        boards:projectItems(first:$boardItems){nodes{project{id number fields(first:$nestedFirst){nodes{
          ... on ProjectV2SingleSelectField{__typename id name options{id name}}
          ... on ProjectV2Field{__typename id name dataType}
        }pageInfo{hasNextPage}}}}}
        blockedBy(first:$first){nodes{...Related}pageInfo{hasNextPage endCursor}}
        project:parent{__typename ...BoardIssue
          subIssues(first:$memberFirst){pageInfo{hasNextPage endCursor} nodes{__typename ...BoardIssue}}
        }
      }}
    }"#,
        board_issue!(),
        related_issue!()
    );

    /// One project's tasks: the sub-issues of the issue that project is, each with a page of
    /// what blocks it.
    ///
    /// Paged at `$memberFirst` — [`MEMBER_PAGE_SIZE`](super::MEMBER_PAGE_SIZE), the most rows
    /// one point buys — rather than at `$first`, because every row carries six connections and
    /// a hundred-row page cost six points to list a project of ten.
    ///
    /// The work this costs is the project's own size. Nothing about it grows as the board
    /// gains projects, or as those projects gain tasks.
    ///
    /// **Why `blockedBy` rides here and on no other page of issues.** What reads a project's
    /// tasks reads their edges next — `project graph` draws them, a copy carries them — and
    /// without them here that is one [`ISSUE_DEPENDENCIES`] per task, so the requests a graph
    /// costs grow with its tasks rather than with the pages of them. Carried here, a task
    /// blocked by no more than `$nestedFirst` issues answers its forward edges from this read,
    /// exactly as an [`ISSUE`] read of it does, and only one blocked by more is asked again.
    /// It adds a connection under each issue of the page — one rate-limit point per page,
    /// and `$nestedFirst` nodes per issue — and is kept off [`SEARCH_ISSUES`], whose pages
    /// answer questions that never read an edge.
    ///
    /// **The project issue itself rides here too**, with the page of what blocks it, because
    /// the reader of a project's tasks reads the project next: a copy writing the project reads
    /// it and its edges before it writes them. Carried on the request that already names the
    /// project, the first page answers that read and the dependency read after it, at the
    /// state GitHub answered in this command, rather than spending one more read on an item
    /// the cache's token cannot vouch for whole. Each sub-issue is kept the same way, so a
    /// copy's read of a task this listing carried is answered by it rather than sent again.
    /// Only a copy's read-before-write takes them; every other read by id reads GitHub.
    pub const SUB_ISSUES: &str = concat!(
        r#"query($id:ID!,$memberFirst:Int!,$after:String,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){
      node(id:$id){__typename ...BoardIssue
        ... on Issue{blockedBy(first:$nestedFirst){nodes{...Related}pageInfo{hasNextPage endCursor}}
          subIssues(first:$memberFirst,after:$after){
          pageInfo{hasNextPage endCursor}
          nodes{__typename ...BoardIssue ... on Issue{
            blockedBy(first:$nestedFirst){nodes{...Related}pageInfo{hasNextPage endCursor}}
          }}
        }}}
    }"#,
        board_issue!(),
        related_issue!()
    );

    /// Everything this source reads about one issue comment, wherever it reaches one.
    ///
    /// A macro for the reason [`board_issue!`] is one: a comment listed, a comment just added
    /// and a comment just edited are handed to one mapper, so they are selected by one
    /// spelling. `author` is `Actor`, which GitHub answers `null` for an account that no
    /// longer exists, and `login` is the one member every kind of actor carries; `__typename`
    /// says whether it is a `Bot`, whose `login` GitHub's own pages spell with `[bot]` after it.
    macro_rules! issue_comment {
        () => {
            "id author{__typename login} createdAt updatedAt body url"
        };
    }

    /// What a read of the board's own `items` selects of each item's content.
    ///
    /// A macro for the reason [`board_item_values!`] is one: [`BOARD`] and [`ORIGIN_LOOKUP`]
    /// both walk `ProjectV2.items` and hand each item to one resolver, so they select its
    /// content by one spelling.
    macro_rules! board_item_content {
        () => { board_item_content!("") };
        ($comments:expr) => {
            concat!(r#" content{
        ... on Issue{__typename id number title body url createdAt updatedAt state stateReason(enableDuplicate:$duplicates) repository{nameWithOwner} parent{id} subIssuesSummary{total} commentCount:comments{totalCount} relationshipEvents:timelineItems(itemTypes:$relationshipEvents){totalCount} labels(first:$nestedFirst){nodes{id name color}pageInfo{hasNextPage}}"#, $comments, r#"}
        ... on PullRequest{__typename id}
        ... on DraftIssue{__typename id title body createdAt updatedAt}
      }"#)
        };
    }

    macro_rules! board_listing {
        ($variables:expr, $comments:expr) => { concat!(
        r#"query($owner:String!,$number:Int!,$first:Int!,$after:String,$nestedFirst:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(), $variables,
        r#"){
      owner:repositoryOwner(login:$owner){
        ... on ProjectV2Owner{projectV2(number:$number){...Board}}
      }
    } fragment Board on ProjectV2 { id title
      fields(first:$nestedFirst){nodes{
        ... on ProjectV2SingleSelectField{__typename id name options{id name}}
        ... on ProjectV2Field{__typename id name dataType}
      }pageInfo{hasNextPage}}
      items(first:$first,after:$after){nodes{id updatedAt "#,
        board_item_values!(),
        board_item_content!($comments),
        r#"} pageInfo{hasNextPage endCursor}}
    }"#
    ) };
    }

    /// Reads the board's fields and one page of its items.
    pub const BOARD: &str = board_listing!("", "");

    /// A board page carrying each issue's first comment page in the same request.
    pub const BOARD_WITH_COMMENTS: &str = concat!(
        board_listing!(
            ",$commentFirst:Int!",
            concat!(
                "comments(first:$commentFirst){nodes{",
                issue_comment!(),
                "}pageInfo{hasNextPage endCursor}} blockedBy(first:$nestedFirst){nodes{...Related}pageInfo{hasNextPage endCursor}}"
            )
        ),
        related_issue!()
    );

    /// Every carrier of one copy origin, by two reads in one request, and nothing else of
    /// the board.
    ///
    /// **`originItems`** is the board's own items narrowed by its own field filter —
    /// `ProjectV2.items(query:)`, which GitHub's schema declares as "Search query for
    /// filtering items" — to those whose `onetaskgraph.origin` text field holds the
    /// qualified id, quoted. It reads the field every carrier already holds, whichever release
    /// wrote it, and matches it exactly: measured on 2026-09-29 against a 394-item board,
    /// the quoted, the unquoted and the bare-value spellings each returned exactly the one
    /// carrier and a prefix of the value returned none. It is `ProjectV2.items`, so it lags a
    /// fresh `addProjectV2ItemById` the way that connection does.
    ///
    /// **`search`** is the board-scoped issue search for the same id as a quoted phrase in
    /// the body, which is where this source mirrors the origin into its metadata slot. GitHub
    /// indexes that comment, and the index catches up with a write in a second or two rather
    /// than in minutes, so it finds a carrier another process wrote that the first read is
    /// still behind on.
    ///
    /// Each connection pages at `$originFirst`, its own small size — see `ORIGIN_PAGE_SIZE`
    /// — and resumes from its own cursor; a connection already walked to its end is resumed
    /// from its last cursor, which answers an empty page. Every candidate either read returns
    /// is confirmed against its own origin field before it is reported, so a token match of
    /// the search or anything else the filter admits never is.
    ///
    /// The root is aliased `originItems` rather than `owner`, so nothing counting the board's
    /// own whole reads counts this one among them.
    pub const ORIGIN_LOOKUP: &str = concat!(
        r#"query($owner:String!,$number:Int!,$filter:String!,$search:String!,$type:SearchType!,$originFirst:Int!,$itemsAfter:String,$searchAfter:String,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){
      originItems:repositoryOwner(login:$owner){
        ... on ProjectV2Owner{projectV2(number:$number){
          items(first:$originFirst,after:$itemsAfter,query:$filter){nodes{id updatedAt "#,
        board_item_values!(),
        board_item_content!(),
        r#"} pageInfo{hasNextPage endCursor}}
        }}
      }
      search(query:$search,type:$type,first:$originFirst,after:$searchAfter){
        pageInfo{hasNextPage endCursor}
        nodes{__typename ...BoardIssue}
      }
    }"#,
        board_issue!()
    );

    /// Every closed item of the board, by the board's own field filter — what
    /// `task status reconcile` reads, and nothing else of the board.
    ///
    /// **`ProjectV2.items(query:)` narrowed by `$filter`**, which this source binds to
    /// `is:closed` alone, so a board whose open items are most of it is not read whole to find
    /// the closed ones. It is narrowed rather than trusted: every item it answers is confirmed
    /// closed, with a reason and a card that contradict each other, from the item node it
    /// already carries, so a filter that also let open items through would cost reads and
    /// write nothing wrong.
    ///
    /// The board's own id and field definitions ride along, as [`BOARD`] carries them, so the
    /// option writes that follow need no read of the board. The root is aliased `closedItems`
    /// rather than `owner`, so nothing counting the board's own whole reads counts this one
    /// among them.
    pub const CLOSED_ITEMS: &str = concat!(
        r#"query($owner:String!,$number:Int!,$first:Int!,$after:String,$filter:String!,$nestedFirst:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){
      closedItems:repositoryOwner(login:$owner){
        ... on ProjectV2Owner{projectV2(number:$number){id
          fields(first:$nestedFirst){nodes{
            ... on ProjectV2SingleSelectField{__typename id name options{id name}}
            ... on ProjectV2Field{__typename id name dataType}
          }pageInfo{hasNextPage}}
          items(first:$first,after:$after,query:$filter){nodes{id updatedAt "#,
        board_item_values!(),
        board_item_content!(),
        r#"} pageInfo{hasNextPage endCursor}}
        }}
      }
    }"#
    );

    /// One alias of [`ISSUES_BY_REFERENCE`]: the issue `$owner<n>/$name<n>#$number<n>` names.
    macro_rules! reference_alias {
        ($n:literal) => {
            concat!(
                "\n      r",
                stringify!($n),
                ":repository(owner:$owner",
                stringify!($n),
                ",name:$name",
                stringify!($n),
                "){issue(number:$number",
                stringify!($n),
                "){...HeldIssue}}"
            )
        };
    }

    /// [`REFERENCE_BATCH`](super::REFERENCE_BATCH) issues at once, each by its repository and
    /// number — the one read `task status complete` makes of a source, and nothing else of
    /// the board: no `items`, no search and no project read.
    ///
    /// **Its own fragment, lighter than `board_issue!`, and that is what makes fifty
    /// references one point.** It selects no labels, and of each board membership only the
    /// `Status` value, by name, with `fieldValueByName` — so each alias asks GitHub for one
    /// connection, its page of memberships, where the whole fragment asks for five. What a
    /// completion writes is a card and, at most, the issue's state, so the labels and the other
    /// field values it leaves out are never what it needs; an item read this way is held for
    /// that write alone and never answers another read. An issue whose entry for this board is
    /// past its first page of memberships is walked with [`ISSUE_BOARD_ITEMS`], exactly as
    /// every other read of an issue recovers one.
    ///
    /// Fixed-size, as [`ISSUE_DETAILS`] is, so there is one document to price: a batch of
    /// fewer references binds the slots it has none for to its last one.
    pub const ISSUES_BY_REFERENCE: &str = concat!(
        r#"query($owner0:String!,$name0:String!,$number0:Int!,$owner1:String!,$name1:String!,$number1:Int!,$owner2:String!,$name2:String!,$number2:Int!,$owner3:String!,$name3:String!,$number3:Int!,$owner4:String!,$name4:String!,$number4:Int!,$owner5:String!,$name5:String!,$number5:Int!,$owner6:String!,$name6:String!,$number6:Int!,$owner7:String!,$name7:String!,$number7:Int!,$owner8:String!,$name8:String!,$number8:Int!,$owner9:String!,$name9:String!,$number9:Int!,$owner10:String!,$name10:String!,$number10:Int!,$owner11:String!,$name11:String!,$number11:Int!,$owner12:String!,$name12:String!,$number12:Int!,$owner13:String!,$name13:String!,$number13:Int!,$owner14:String!,$name14:String!,$number14:Int!,$owner15:String!,$name15:String!,$number15:Int!,$owner16:String!,$name16:String!,$number16:Int!,$owner17:String!,$name17:String!,$number17:Int!,$owner18:String!,$name18:String!,$number18:Int!,$owner19:String!,$name19:String!,$number19:Int!,$owner20:String!,$name20:String!,$number20:Int!,$owner21:String!,$name21:String!,$number21:Int!,$owner22:String!,$name22:String!,$number22:Int!,$owner23:String!,$name23:String!,$number23:Int!,$owner24:String!,$name24:String!,$number24:Int!,$owner25:String!,$name25:String!,$number25:Int!,$owner26:String!,$name26:String!,$number26:Int!,$owner27:String!,$name27:String!,$number27:Int!,$owner28:String!,$name28:String!,$number28:Int!,$owner29:String!,$name29:String!,$number29:Int!,$owner30:String!,$name30:String!,$number30:Int!,$owner31:String!,$name31:String!,$number31:Int!,$owner32:String!,$name32:String!,$number32:Int!,$owner33:String!,$name33:String!,$number33:Int!,$owner34:String!,$name34:String!,$number34:Int!,$owner35:String!,$name35:String!,$number35:Int!,$owner36:String!,$name36:String!,$number36:Int!,$owner37:String!,$name37:String!,$number37:Int!,$owner38:String!,$name38:String!,$number38:Int!,$owner39:String!,$name39:String!,$number39:Int!,$owner40:String!,$name40:String!,$number40:Int!,$owner41:String!,$name41:String!,$number41:Int!,$owner42:String!,$name42:String!,$number42:Int!,$owner43:String!,$name43:String!,$number43:Int!,$owner44:String!,$name44:String!,$number44:Int!,$owner45:String!,$name45:String!,$number45:Int!,$owner46:String!,$name46:String!,$number46:Int!,$owner47:String!,$name47:String!,$number47:Int!,$owner48:String!,$name48:String!,$number48:Int!,$owner49:String!,$name49:String!,$number49:Int!,$boardItems:Int!,$statusField:String!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){"#,
        reference_alias!(0),
        reference_alias!(1),
        reference_alias!(2),
        reference_alias!(3),
        reference_alias!(4),
        reference_alias!(5),
        reference_alias!(6),
        reference_alias!(7),
        reference_alias!(8),
        reference_alias!(9),
        reference_alias!(10),
        reference_alias!(11),
        reference_alias!(12),
        reference_alias!(13),
        reference_alias!(14),
        reference_alias!(15),
        reference_alias!(16),
        reference_alias!(17),
        reference_alias!(18),
        reference_alias!(19),
        reference_alias!(20),
        reference_alias!(21),
        reference_alias!(22),
        reference_alias!(23),
        reference_alias!(24),
        reference_alias!(25),
        reference_alias!(26),
        reference_alias!(27),
        reference_alias!(28),
        reference_alias!(29),
        reference_alias!(30),
        reference_alias!(31),
        reference_alias!(32),
        reference_alias!(33),
        reference_alias!(34),
        reference_alias!(35),
        reference_alias!(36),
        reference_alias!(37),
        reference_alias!(38),
        reference_alias!(39),
        reference_alias!(40),
        reference_alias!(41),
        reference_alias!(42),
        reference_alias!(43),
        reference_alias!(44),
        reference_alias!(45),
        reference_alias!(46),
        reference_alias!(47),
        reference_alias!(48),
        reference_alias!(49),
        "\n    }",
        r#" fragment HeldIssue on Issue{__typename id number title body url createdAt updatedAt state stateReason(enableDuplicate:$duplicates) repository{nameWithOwner} parent{id} subIssuesSummary{total} commentCount:comments{totalCount} relationshipEvents:timelineItems(itemTypes:$relationshipEvents){totalCount}
      projectItems(first:$boardItems){nodes{id updatedAt project{id number}
        status:fieldValueByName(name:$statusField){... on ProjectV2ItemFieldSingleSelectValue{name field{
          ... on ProjectV2SingleSelectField{id name options{id name}}
        }}}
      }pageInfo{hasNextPage endCursor}}}"#
    );

    /// The board's own id and field definitions, and not one of its items.
    ///
    /// What a write needs of the board when the item it writes does not say: the id a field
    /// write and `addProjectV2ItemById` address, and the definitions of the `Status` and
    /// origin fields. It selects no `items`, so what it costs is the board's field list
    /// however many items the board holds — and it decides nothing about which items those
    /// are, which is the question a read of one item by its own id answers instead.
    ///
    /// The root is aliased `boardFields` rather than `owner`, so nothing counting the
    /// board's item reads by their root counts this one among them.
    pub const BOARD_FIELDS: &str = r#"query($owner:String!,$number:Int!,$nestedFirst:Int!){
      boardFields:repositoryOwner(login:$owner){
        ... on ProjectV2Owner{projectV2(number:$number){id
          fields(first:$nestedFirst){nodes{
            ... on ProjectV2SingleSelectField{__typename id name options{id name}}
            ... on ProjectV2Field{__typename id name dataType}
          }pageInfo{hasNextPage}}
        }}
      }
    }"#;

    /// One board draft by its own node id, with the board item it sits in.
    ///
    /// A draft is not an issue, so [`ISSUE`] reaches it and reads nothing of it; this is the
    /// second read that answers it. `DraftIssue.projectV2Items` names the board item a draft
    /// is — GitHub links a draft to one item — with the same [`board_item_values!`] the
    /// issue fragment reads, so a draft reached by id resolves through the same resolver a
    /// board listing hands it to, and nothing has to list the board to find one.
    pub const DRAFT: &str = concat!(
        r#"query($id:ID!,$nestedFirst:Int!,$boardItems:Int!){
      node(id:$id){__typename ... on DraftIssue{id title body createdAt updatedAt
        projectV2Items(first:$boardItems){nodes{id project{id number}
        "#,
        board_item_values!(),
        r#"}pageInfo{hasNextPage endCursor}}}}
    }"#
    );

    /// One issue's board memberships alone, walked past the page a read of it carried.
    ///
    /// The recovery read behind [`GitHubProjectsSource::resolve_issue`](super::GitHubProjectsSource):
    /// every document above carries a *page* of `Issue.projectItems`, and an issue on more
    /// boards than that page holds may have this board's entry past its end. This asks that
    /// one issue for its memberships and nothing else — the caller already holds the issue —
    /// so an answer of "this board does not hold it" is only ever given about a connection
    /// read to exhaustion.
    ///
    /// It selects the board item's id, its project number and the same
    /// [`board_item_values!`] the fragment does, because what it produces is handed to the
    /// very same resolver: an issue recovered this way reports the same title, the same
    /// status, the same labels and the same qualified id as one whose entry was on the
    /// page.
    ///
    /// `$first` rather than `$boardItems`: this document reads one issue, so nothing
    /// multiplies through it and the membership connection can be walked at
    /// [`MAX_PAGE_SIZE`](super::MAX_PAGE_SIZE) — which is what keeps the recovery to one
    /// further request for any issue a person really keeps.
    pub const ISSUE_BOARD_ITEMS: &str = concat!(
        r#"query($id:ID!,$first:Int!,$after:String,$nestedFirst:Int!){
      node(id:$id){
        ... on Issue{projectItems(first:$first,after:$after){
          nodes{id updatedAt project{id number}
        "#,
        board_item_values!(),
        r#"}
          pageInfo{hasNextPage endCursor}}}
      }
    }"#
    );
    /// Whether the board's Project is public — half of what decides whether a write here can
    /// be read by anybody. Needs the `read:project` scope.
    pub const PROJECT_VISIBILITY: &str = r#"query($owner:String!,$number:Int!){visibility:repositoryOwner(login:$owner){... on ProjectV2Owner{projectV2(number:$number){public}}}}"#;
    /// Resolves the configured repository's node id, which creating an issue requires.
    pub const REPOSITORY: &str = r#"query($owner:String!,$name:String!){repository(owner:$owner,name:$name){id nameWithOwner}}"#;
    /// What creating an issue needs and has not read yet: the board's own id and field
    /// definitions, as [`BOARD_FIELDS`] reads them, and the node id of the repository the
    /// issue is created in, as [`REPOSITORY`] reads it — in one request.
    ///
    /// Sent at the point a create knows which repository it is for, when neither half is
    /// already known to this process; a create needing only one of them sends that one's own
    /// document. Neither half is kept past the process: a field's option ids are re-minted by
    /// `sources fields --apply`, so a copy of them held between runs would write the wrong
    /// status.
    pub const CREATION_CONTEXT: &str = r#"query($owner:String!,$number:Int!,$nestedFirst:Int!,$repositoryOwner:String!,$repositoryName:String!){
      boardFields:repositoryOwner(login:$owner){
        ... on ProjectV2Owner{projectV2(number:$number){id
          fields(first:$nestedFirst){nodes{
            ... on ProjectV2SingleSelectField{__typename id name options{id name}}
            ... on ProjectV2Field{__typename id name dataType}
          }pageInfo{hasNextPage}}
        }}
      }
      repository(owner:$repositoryOwner,name:$repositoryName){id nameWithOwner}
    }"#;
    /// Reads both dependency directions for one issue, with each far end's own kind — and
    /// the issue's own body, which is where an edge to another source is recorded, so that
    /// half of a dependency read needs no second read of the issue or of the board.
    pub const ISSUE_DEPENDENCIES: &str = concat!(
        r#"query($id:ID!,$first:Int!,$after:String){node(id:$id){__typename
      ... on Issue{body
        blockedBy(first:$first,after:$after){nodes{...Related}pageInfo{hasNextPage endCursor}}
        blocking(first:$first,after:$after){nodes{...Related}pageInfo{hasNextPage endCursor}}
      }}}"#,
        related_issue!()
    );
    /// What one item depends on: [`ISSUE_DEPENDENCIES`] narrowed to `blockedBy`, so a
    /// dependency read asks GitHub for the one direction it answers. The issue's `body` rides
    /// along because its metadata slot is where an edge to another source is recorded — and a
    /// board draft's `body` too, with the one board item GitHub links a draft to and, aliased
    /// `board`, this board's own id, so a draft's recorded edges are read off this same answer
    /// and only when that item is this board's: no draft read and no board read follow it.
    ///
    /// The draft's board item is asked for under a root of its own, aliased `draft`, rather
    /// than beside its `body` under `node`. GitHub counts the type-conditional fragments of
    /// one selection as the largest of them — `blockedBy`'s 100 alone, where this workspace's
    /// calculation adds the draft's 3 — while separate roots are summed by both, so this is
    /// the spelling whose node count GitHub and the calculation agree on. Both price it at
    /// one point; the credentialed lane's reconciliation is what holds the two to that.
    pub const ISSUE_DEPENDS_ON: &str = concat!(
        r#"query($id:ID!,$first:Int!,$after:String,$owner:String!,$number:Int!,$boardItems:Int!){
      node(id:$id){__typename
      ... on Issue{body
        blockedBy(first:$first,after:$after){nodes{...Related}pageInfo{hasNextPage endCursor}}
      }
      ... on DraftIssue{id body}}
      draft:node(id:$id){... on DraftIssue{projectV2Items(first:$boardItems){nodes{project{id number}}pageInfo{hasNextPage}}}}
      board:repositoryOwner(login:$owner){... on ProjectV2Owner{projectV2(number:$number){id}}}}"#,
        related_issue!()
    );
    /// What depends on one issue: [`ISSUE_DEPENDENCIES`] with its `blocking` alone. No `body`:
    /// the reverse of a recorded edge is derived from the far end, never written down here.
    pub const ISSUE_DEPENDED_ON_BY: &str = concat!(
        r#"query($id:ID!,$first:Int!,$after:String){node(id:$id){__typename
      ... on Issue{
        blocking(first:$first,after:$after){nodes{...Related}pageInfo{hasNextPage endCursor}}
      }}}"#,
        related_issue!()
    );
    /// Creates one issue. Outside a copy it is created on no board, and [`ADD_TO_BOARD`] files
    /// it; inside one, its input names the board in `projectV2Ids` and its project in
    /// `parentIssueId`, so this one request files it under both — see
    /// [`GitHubProjectsSource::create_filed_issue`](super::GitHubProjectsSource).
    pub const CREATE_ISSUE: &str =
        r#"mutation($input:CreateIssueInput!){createIssue(input:$input){issue{id number url}}}"#;
    /// Puts an existing issue on the configured board.
    pub const ADD_TO_BOARD: &str = r#"mutation($input:AddProjectV2ItemByIdInput!){addProjectV2ItemById(input:$input){item{id}}}"#;
    /// Updates an issue's visible fields and its open or closed state in one call.
    pub const UPDATE_ISSUE: &str =
        r#"mutation($input:UpdateIssueInput!){updateIssue(input:$input){issue{id}}}"#;
    /// Updates an existing draft's user-visible fields.
    pub const UPDATE_DRAFT: &str = r#"mutation($input:UpdateProjectV2DraftIssueInput!){updateProjectV2DraftIssue(input:$input){draftIssue{id}}}"#;
    /// Updates a text or single-select value on one project item.
    pub const UPDATE_FIELD: &str = r#"mutation($input:UpdateProjectV2ItemFieldValueInput!,$readPriority:Boolean!,$priorityName:String!){updateProjectV2ItemFieldValue(input:$input){projectV2Item{id fieldValueByName(name:$priorityName) @include(if:$readPriority){... on ProjectV2ItemFieldSingleSelectValue{name field{... on ProjectV2SingleSelectField{id name options{id name}}}}}}}}"#;
    /// Writes up to six board fields and clears up to three in one ordered mutation: every
    /// aliased field is included by its own boolean, as [`FIELD_WRITE_SLOTS`](super::FIELD_WRITE_SLOTS)
    /// and [`FIELD_CLEAR_SLOTS`](super::FIELD_CLEAR_SLOTS) name them, writes before clears.
    pub const UPDATE_FIELDS: &str = r#"mutation($input:UpdateProjectV2ItemFieldValueInput!,$second:UpdateProjectV2ItemFieldValueInput!,$third:UpdateProjectV2ItemFieldValueInput!,$fourth:UpdateProjectV2ItemFieldValueInput!,$fifth:UpdateProjectV2ItemFieldValueInput!,$sixth:UpdateProjectV2ItemFieldValueInput!,$clear:ClearProjectV2ItemFieldValueInput!,$clearSecond:ClearProjectV2ItemFieldValueInput!,$clearThird:ClearProjectV2ItemFieldValueInput!,$writeFirst:Boolean!,$writeSecond:Boolean!,$writeThird:Boolean!,$writeFourth:Boolean!,$writeFifth:Boolean!,$writeSixth:Boolean!,$writeClear:Boolean!,$writeClearSecond:Boolean!,$writeClearThird:Boolean!){updateProjectV2ItemFieldValue(input:$input) @include(if:$writeFirst){projectV2Item{id}} second:updateProjectV2ItemFieldValue(input:$second) @include(if:$writeSecond){projectV2Item{id}} third:updateProjectV2ItemFieldValue(input:$third) @include(if:$writeThird){projectV2Item{id}} fourth:updateProjectV2ItemFieldValue(input:$fourth) @include(if:$writeFourth){projectV2Item{id}} fifth:updateProjectV2ItemFieldValue(input:$fifth) @include(if:$writeFifth){projectV2Item{id}} sixth:updateProjectV2ItemFieldValue(input:$sixth) @include(if:$writeSixth){projectV2Item{id}} cleared:clearProjectV2ItemFieldValue(input:$clear) @include(if:$writeClear){projectV2Item{id}} clearedSecond:clearProjectV2ItemFieldValue(input:$clearSecond) @include(if:$writeClearSecond){projectV2Item{id}} clearedThird:clearProjectV2ItemFieldValue(input:$clearThird) @include(if:$writeClearThird){projectV2Item{id}}}"#;
    /// Clears one project item's value of one field, which is what a `none` priority is.
    pub const CLEAR_FIELD: &str = r#"mutation($input:ClearProjectV2ItemFieldValueInput!,$readPriority:Boolean!,$priorityName:String!){clearProjectV2ItemFieldValue(input:$input){projectV2Item{id fieldValueByName(name:$priorityName) @include(if:$readPriority){... on ProjectV2ItemFieldSingleSelectValue{name field{... on ProjectV2SingleSelectField{id name options{id name}}}}}}}}"#;
    /// Creates one field: a single-select one with its options, or a text one for a projected
    /// metadata value. Only the guarded field setup may use this document, and only for a
    /// field the board lacks.
    pub const CREATE_FIELD: &str = r#"mutation($input:CreateProjectV2FieldInput!){createProjectV2Field(input:$input){projectV2Field{... on ProjectV2SingleSelectField{id name options{id name color description}}}}}"#;
    /// Replaces a single-select field's options. Only the guarded field setup — the
    /// `status-options` and `fields` operations — may use this document, because GitHub
    /// treats the input as the complete option list.
    pub const STATUS_OPTIONS_UPDATE: &str = r#"mutation($input:UpdateProjectV2FieldInput!){updateProjectV2Field(input:$input){projectV2Field{... on ProjectV2SingleSelectField{id options{id name color description}}}}}"#;
    /// A fresh snapshot of the Status field and every board item's assignment.
    pub const STATUS_OPTIONS_SNAPSHOT: &str = r#"query($owner:String!,$number:Int!,$first:Int!,$after:String,$nestedFirst:Int!){owner:repositoryOwner(login:$owner){... on ProjectV2Owner{projectV2(number:$number){id fields(first:$nestedFirst){nodes{... on ProjectV2SingleSelectField{id name options{id name color description}}}pageInfo{hasNextPage}} items(first:$first,after:$after){nodes{id fieldValues(first:$nestedFirst){nodes{... on ProjectV2ItemFieldSingleSelectValue{name optionId field{... on ProjectV2SingleSelectField{id name}}}}pageInfo{hasNextPage}}}pageInfo{hasNextPage endCursor}}}}}}"#;
    /// Files one issue under another as a sub-issue, which is what project membership is.
    pub const ADD_SUB_ISSUE: &str =
        r#"mutation($input:AddSubIssueInput!){addSubIssue(input:$input){issue{id} subIssue{id}}}"#;
    /// Takes one issue back out of its parent.
    pub const REMOVE_SUB_ISSUE: &str = r#"mutation($input:RemoveSubIssueInput!){removeSubIssue(input:$input){issue{id} subIssue{id}}}"#;
    /// Adds GitHub's native issue blocked-by relationship.
    pub const ADD_BLOCKED_BY: &str = r#"mutation($input:AddBlockedByInput!){addBlockedBy(input:$input){issue{id} blockingIssue{id}}}"#;
    /// One origin of [`ORIGIN_LOOKUPS`]: [`ORIGIN_LOOKUP`]'s two connections for the origin the
    /// `$filter<n>` and `$search<n>` of slot `<n>` name, sent only while `$ask<n>` holds.
    macro_rules! origin_lookup_alias {
        ($n:literal) => {
            concat!(
                "\n      o",
                stringify!($n),
                ":repositoryOwner(login:$owner) @include(if:$ask",
                stringify!($n),
                "){... on ProjectV2Owner{projectV2(number:$number){items(first:$lookupFirst,query:$filter",
                stringify!($n),
                "){nodes{id updatedAt ",
                board_item_values!(),
                board_item_content!(),
                "} pageInfo{hasNextPage endCursor}}}}}\n      s",
                stringify!($n),
                ":search(query:$search",
                stringify!($n),
                ",type:$type,first:$lookupFirst) @include(if:$ask",
                stringify!($n),
                "){pageInfo{hasNextPage endCursor} nodes{__typename ...BoardIssue}}"
            )
        };
    }

    /// [`ORIGIN_LOOKUPS`], with `$extra_variables` declared after its own and `$extra_fields`
    /// selected beside them: one spelling of the batched lookup for the documents that send it.
    macro_rules! origin_lookups {
        ($extra_variables:literal, $extra_fields:literal) => {
            concat!(
                r#"query($owner:String!,$number:Int!,$type:SearchType!,$lookupFirst:Int!,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!,$repositoryOwner:String!,$repositoryName:String!,$filter0:String!,$filter1:String!,$filter2:String!,$filter3:String!,$filter4:String!,$filter5:String!,$filter6:String!,$filter7:String!,$filter8:String!,$filter9:String!,$search0:String!,$search1:String!,$search2:String!,$search3:String!,$search4:String!,$search5:String!,$search6:String!,$search7:String!,$search8:String!,$search9:String!,$ask0:Boolean!,$ask1:Boolean!,$ask2:Boolean!,$ask3:Boolean!,$ask4:Boolean!,$ask5:Boolean!,$ask6:Boolean!,$ask7:Boolean!,$ask8:Boolean!,$ask9:Boolean!"#,
                relationship_events_variable!(),
                $extra_variables,
                r#"){
      boardFields:repositoryOwner(login:$owner){
        ... on ProjectV2Owner{projectV2(number:$number){id
          fields(first:$nestedFirst){nodes{
            ... on ProjectV2SingleSelectField{__typename id name options{id name}}
            ... on ProjectV2Field{__typename id name dataType}
          }pageInfo{hasNextPage}}
        }}
      }
      repository(owner:$repositoryOwner,name:$repositoryName){id databaseId nameWithOwner}"#,
                $extra_fields,
                origin_lookup_alias!(0),
                origin_lookup_alias!(1),
                origin_lookup_alias!(2),
                origin_lookup_alias!(3),
                origin_lookup_alias!(4),
                origin_lookup_alias!(5),
                origin_lookup_alias!(6),
                origin_lookup_alias!(7),
                origin_lookup_alias!(8),
                origin_lookup_alias!(9),
                "\n    }",
                board_issue!()
            )
        };
    }

    /// The first page of [`ORIGIN_LOOKUP`] for up to [`ORIGIN_BATCH`](super::ORIGIN_BATCH)
    /// origins a copy is about to look for, and everything creating an item then reads —
    /// the board's own id and field definitions, as [`CREATION_CONTEXT`] reads them, and the node
    /// id and the numeric id of the repository an issue is created in — in one request.
    ///
    /// What a copy reads before it writes anything, so that finding none of its records on the
    /// board and creating every one costs this read and the writes. An origin whose page says
    /// there is more is left to its own [`ORIGIN_LOOKUP`]. The repository's numeric id is what an
    /// image upload names, so a copy carrying assets needs no `GET /repos/{owner}/{repo}` either.
    pub const ORIGIN_LOOKUPS: &str = origin_lookups!("", "");

    /// [`ORIGIN_LOOKUPS`], with a page of the members of the project `$project` beside it, each
    /// trimmed to what classifying it reads — its id, its parent, its body and its repository,
    /// and no connection under it.
    ///
    /// What a document copy sends where the public boundary screens its write: the document is
    /// held to who may read its project, which is its members' classification as much as the
    /// project's own, and this request carries both the project — among the origins it asks —
    /// and its members. GitHub's own dry run prices it at the one point [`ORIGIN_LOOKUPS`]
    /// costs alone (`docs/measurements/board-copy-gaps-2026-10-24/coalesced-members-dry-run`).
    pub const ORIGIN_LOOKUPS_WITH_MEMBERS: &str = origin_lookups!(
        ",$project:ID!,$memberFirst:Int!",
        "\n      members:node(id:$project){... on Issue{subIssues(first:$memberFirst){pageInfo{hasNextPage endCursor} nodes{id body parent{id} repository{nameWithOwner}}}}}"
    );

    /// One slot `<n>` of a copy's aliased writes: `$<prefix><n>` is its input and `$send<n>`
    /// whether it is sent.
    macro_rules! write_slot {
        ($prefix:literal, $field:literal, $selection:literal, $n:literal) => {
            concat!(
                " ",
                $prefix,
                stringify!($n),
                ":",
                $field,
                "(input:$",
                $prefix,
                stringify!($n),
                ") @include(if:$send",
                stringify!($n),
                "){",
                $selection,
                "}"
            )
        };
    }

    /// The variables of a copy's aliased writes: eleven inputs of `$type` named `$<prefix><n>`,
    /// and eleven `$send<n>`.
    macro_rules! write_variables {
        ($prefix:literal, $type:literal) => {
            concat!(
                "mutation(",
                "$",
                $prefix,
                "0:",
                $type,
                "!,$",
                $prefix,
                "1:",
                $type,
                "!,$",
                $prefix,
                "2:",
                $type,
                "!,$",
                $prefix,
                "3:",
                $type,
                "!,$",
                $prefix,
                "4:",
                $type,
                "!,$",
                $prefix,
                "5:",
                $type,
                "!,$",
                $prefix,
                "6:",
                $type,
                "!,$",
                $prefix,
                "7:",
                $type,
                "!,$",
                $prefix,
                "8:",
                $type,
                "!,$",
                $prefix,
                "9:",
                $type,
                "!,$",
                $prefix,
                "10:",
                $type,
                "!,",
                "$send0:Boolean!,$send1:Boolean!,$send2:Boolean!,$send3:Boolean!,$send4:Boolean!,",
                "$send5:Boolean!,$send6:Boolean!,$send7:Boolean!,$send8:Boolean!,$send9:Boolean!,",
                "$send10:Boolean!){"
            )
        };
    }

    /// A copy's aliased writes of one kind: [`MUTATION_BATCH`](super::MUTATION_BATCH) slots of
    /// `$field`, each sent only while its `$send<n>` holds.
    macro_rules! write_batch {
        ($prefix:literal, $type:literal, $field:literal, $selection:literal) => {
            concat!(
                write_variables!($prefix, $type),
                write_slot!($prefix, $field, $selection, 0),
                write_slot!($prefix, $field, $selection, 1),
                write_slot!($prefix, $field, $selection, 2),
                write_slot!($prefix, $field, $selection, 3),
                write_slot!($prefix, $field, $selection, 4),
                write_slot!($prefix, $field, $selection, 5),
                write_slot!($prefix, $field, $selection, 6),
                write_slot!($prefix, $field, $selection, 7),
                write_slot!($prefix, $field, $selection, 8),
                write_slot!($prefix, $field, $selection, 9),
                write_slot!($prefix, $field, $selection, 10),
                "}"
            )
        };
    }

    /// Creates up to [`MUTATION_BATCH`](super::MUTATION_BATCH) issues in one request, each filed on
    /// the board and under its project by the same `createIssue` — its `projectV2Ids` and
    /// `parentIssueId` — as [`CREATE_ISSUE`] creates one.
    pub const CREATE_ISSUES: &str = write_batch!(
        "create",
        "CreateIssueInput",
        "createIssue",
        "issue{id number url}"
    );

    /// One issue of [`CREATED_BOARD_ITEMS`]: the board items the issue `$id<n>` names is on, with
    /// the `Status` value of each, read only while `$read<n>` holds.
    macro_rules! created_item_alias {
        ($n:literal) => {
            concat!(
                "\n      i",
                stringify!($n),
                ":node(id:$id",
                stringify!($n),
                ") @include(if:$read",
                stringify!($n),
                "){... on Issue{id projectItems(first:$boardItems){nodes{id project{id} status:fieldValueByName(name:$statusName){... on ProjectV2ItemFieldSingleSelectValue{name}}}}}}"
            )
        };
    }

    /// The board item of each issue a copy created, up to
    /// [`COPY_DETAIL_BATCH`](super::COPY_DETAIL_BATCH) at a time, with its `Status` as the board
    /// set it: what `createIssue`'s `projectV2Ids` files the issue as, which GitHub adds after
    /// answering the create, so its id comes from a read made once every create has been sent.
    pub const CREATED_BOARD_ITEMS: &str = concat!(
        r#"query($id0:ID!,$id1:ID!,$id2:ID!,$id3:ID!,$id4:ID!,$id5:ID!,$id6:ID!,$id7:ID!,$id8:ID!,$id9:ID!,$id10:ID!,$id11:ID!,$id12:ID!,$id13:ID!,$id14:ID!,$id15:ID!,$id16:ID!,$id17:ID!,$id18:ID!,$id19:ID!,$id20:ID!,$read0:Boolean!,$read1:Boolean!,$read2:Boolean!,$read3:Boolean!,$read4:Boolean!,$read5:Boolean!,$read6:Boolean!,$read7:Boolean!,$read8:Boolean!,$read9:Boolean!,$read10:Boolean!,$read11:Boolean!,$read12:Boolean!,$read13:Boolean!,$read14:Boolean!,$read15:Boolean!,$read16:Boolean!,$read17:Boolean!,$read18:Boolean!,$read19:Boolean!,$read20:Boolean!,$boardItems:Int!,$statusName:String!){"#,
        created_item_alias!(0),
        created_item_alias!(1),
        created_item_alias!(2),
        created_item_alias!(3),
        created_item_alias!(4),
        created_item_alias!(5),
        created_item_alias!(6),
        created_item_alias!(7),
        created_item_alias!(8),
        created_item_alias!(9),
        created_item_alias!(10),
        created_item_alias!(11),
        created_item_alias!(12),
        created_item_alias!(13),
        created_item_alias!(14),
        created_item_alias!(15),
        created_item_alias!(16),
        created_item_alias!(17),
        created_item_alias!(18),
        created_item_alias!(19),
        created_item_alias!(20),
        "\n    }"
    );

    /// The variables and the aliases of [`WRITE_BATCH`](super::WRITE_BATCH) slots: `$<prefix><n>`
    /// is a slot's input and `$<send><n>` whether it is sent, for the slot numbers given.
    macro_rules! wide_variables {
        ($prefix:literal, $type:literal, $send:literal; $($n:literal)+) => {
            concat!($("$", $prefix, stringify!($n), ":", $type, "!,$", $send, stringify!($n), ":Boolean!,",)+)
        };
    }
    macro_rules! wide_slots {
        ($prefix:literal, $field:literal, $selection:literal, $send:literal; $($n:literal)+) => {
            concat!($(
                " ", $prefix, stringify!($n), ":", $field, "(input:$", $prefix, stringify!($n),
                ") @include(if:$", $send, stringify!($n), "){", $selection, "}",
            )+)
        };
    }

    /// Up to [`WRITE_BATCH`](super::WRITE_BATCH) board field values a copy writes on the items it
    /// created, each on its own item, in one ordered request.
    pub const WRITE_FIELD_VALUES: &str = concat!(
        "mutation(",
        wide_variables!("field", "UpdateProjectV2ItemFieldValueInput", "send"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18),
        "){",
        wide_slots!("field", "updateProjectV2ItemFieldValue", "projectV2Item{id}", "send"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18),
        "}"
    );

    /// A copy's board field values and its blocked-by edges together, up to
    /// [`WRITE_BATCH`](super::WRITE_BATCH) of the two kinds between them, in one ordered request:
    /// every field value its `$sendField<n>` sends, then every edge its `$sendBlock<n>` sends.
    /// What a copy sends when one request of its writes holds both kinds; one holding a single
    /// kind is [`WRITE_FIELD_VALUES`] or [`ADD_BLOCKED_BYS`].
    pub const WRITE_FIELDS_AND_EDGES: &str = concat!(
        "mutation(",
        wide_variables!("field", "UpdateProjectV2ItemFieldValueInput", "sendField"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18),
        wide_variables!("block", "AddBlockedByInput", "sendBlock"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18),
        "){",
        wide_slots!("field", "updateProjectV2ItemFieldValue", "projectV2Item{id}", "sendField"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18),
        wide_slots!("block", "addBlockedBy", "issue{id} blockingIssue{id}", "sendBlock"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18),
        "}"
    );

    /// Closes up to [`MUTATION_BATCH`](super::MUTATION_BATCH) issues a copy created at a closing
    /// status, each with its own `stateInput` and nothing else, once its option is selected.
    pub const CLOSE_ISSUES: &str =
        write_batch!("close", "UpdateIssueInput", "updateIssue", "issue{id}");

    /// Updates up to [`MUTATION_BATCH`](super::MUTATION_BATCH) existing issues a copy rewrites,
    /// each with the members its own [`UPDATE_ISSUE`] would have sent, in one request.
    pub const UPDATE_ISSUES: &str =
        write_batch!("update", "UpdateIssueInput", "updateIssue", "issue{id}");

    /// Up to [`WRITE_BATCH`](super::WRITE_BATCH) native blocked-by relationships a copy adds to the
    /// issues it created, in one request.
    pub const ADD_BLOCKED_BYS: &str = concat!(
        "mutation(",
        wide_variables!("block", "AddBlockedByInput", "send"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18),
        "){",
        wide_slots!("block", "addBlockedBy", "issue{id} blockingIssue{id}", "send"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18),
        "}"
    );

    /// Removes one native issue blocked-by relationship.
    pub const REMOVE_BLOCKED_BY: &str = r#"mutation($input:RemoveBlockedByInput!){removeBlockedBy(input:$input){issue{id} blockingIssue{id}}}"#;
    /// Deletes one issue, which takes its board item with it.
    ///
    /// The engine sends this in one situation only: undoing a copy that could not finish,
    /// over the items that same copy created. Deleting the issue removes the board item
    /// too, so there is no second `deleteProjectV2Item` to keep in step with it.
    pub const DELETE_ISSUE: &str =
        r#"mutation($input:DeleteIssueInput!){deleteIssue(input:$input){repository{id}}}"#;
    /// Deletes up to [`DELETE_BATCH`](super::DELETE_BATCH) issues in one request, each as
    /// [`DELETE_ISSUE`] deletes one: what a write that created several issues and was refused
    /// part way sends to take them back.
    pub const DELETE_ISSUES: &str = concat!(
        "mutation(",
        wide_variables!("delete", "DeleteIssueInput", "send"; 0 1 2 3 4),
        "){",
        wide_slots!("delete", "deleteIssue", "repository{id}", "send"; 0 1 2 3 4),
        "}"
    );

    /// One task's comments: a page of its issue's own `comments` connection.
    ///
    /// **No `orderBy`, and that is what makes the page oldest first.** GitHub's only
    /// `IssueCommentOrder` field is `UPDATED_AT`, which would move a comment to the end of the
    /// list every time somebody edited it; left unordered the connection answers in the order
    /// the comments were written, which is the order GitHub documents for the same collection
    /// over REST — ascending id. Nothing multiplies through it, so `$first` is the whole of its
    /// node count and the caller's own page size is pushed straight down.
    pub const ISSUE_COMMENTS: &str = concat!(
        r#"query($id:ID!,$first:Int!,$after:String){node(id:$id){__typename ... on Issue{comments(first:$first,after:$after){nodes{"#,
        issue_comment!(),
        r#"}pageInfo{hasNextPage endCursor}}}}}"#
    );
    /// One issue by its own node id, with a page of its comments: what `task show` and a
    /// comment listing read, in one request.
    ///
    /// [`ISSUE`] and [`ISSUE_COMMENTS`] in one document, rather than one then the other. The
    /// comments are selected here and **not** on the shared `board_issue!` fragment, which
    /// [`SEARCH_ISSUES`] and [`SUB_ISSUES`] nest under a page of a hundred issues: a comment
    /// connection there would multiply through both of those documents' price, and neither
    /// needs one.
    pub const ISSUE_DETAIL: &str = concat!(
        r#"query($id:ID!,$first:Int!,$after:String,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){
      node(id:$id){__typename ...BoardIssue ... on Issue{comments(first:$first,after:$after){nodes{"#,
        issue_comment!(),
        r#"}pageInfo{hasNextPage endCursor}}}}
    }"#,
        board_issue!()
    );

    /// One alias of [`ISSUE_DETAILS`]: the item a batch's `$id<n>` names, with the first
    /// page of its comments when `$comments` asks for them.
    macro_rules! issue_details_alias {
        ($n:literal) => {
            concat!(
                "\n      i",
                stringify!($n),
                ":node(id:$id",
                stringify!($n),
                "){__typename ...BoardIssue ... on Issue{comments(first:$first) @include(if:$comments){nodes{",
                issue_comment!(),
                "}pageInfo{hasNextPage endCursor}}}}"
            )
        };
    }

    /// [`ISSUE_DETAIL`] for [`DETAIL_BATCH`](super::DETAIL_BATCH) items at once, each by its
    /// own node id, as one fixed-size document of aliased `node(id:)` fields.
    ///
    /// **Aliased `node(id:)` rather than `nodes(ids:)`, and that is what keeps its price
    /// honest.** The `github-graphql-node-count` model this workspace prices with treats a
    /// field that supplies neither `first` nor `last` as free, and `nodes(ids:)` supplies
    /// neither — so every connection under it would be priced at nothing and the pin in
    /// `tests/point_cost.rs` would understate what GitHub charges. Each alias here is the
    /// one-item read the model already prices, so the batch costs what its aliases cost.
    ///
    /// **Fixed-size, so there is one document to price.** A batch of fewer items binds the
    /// slots it has no item for to the last item it does, and reads that item again; the
    /// price is the document's, whatever its variables, so a short batch costs what a full
    /// one does and nothing more.
    pub const ISSUE_DETAILS: &str = concat!(
        r#"query($id0:ID!,$id1:ID!,$id2:ID!,$id3:ID!,$id4:ID!,$id5:ID!,$id6:ID!,$id7:ID!,$id8:ID!,$id9:ID!,$id10:ID!,$id11:ID!,$id12:ID!,$id13:ID!,$id14:ID!,$id15:ID!,$id16:ID!,$id17:ID!,$id18:ID!,$id19:ID!,$id20:ID!,$id21:ID!,$id22:ID!,$id23:ID!,$first:Int!,$comments:Boolean!,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){"#,
        issue_details_alias!(0),
        issue_details_alias!(1),
        issue_details_alias!(2),
        issue_details_alias!(3),
        issue_details_alias!(4),
        issue_details_alias!(5),
        issue_details_alias!(6),
        issue_details_alias!(7),
        issue_details_alias!(8),
        issue_details_alias!(9),
        issue_details_alias!(10),
        issue_details_alias!(11),
        issue_details_alias!(12),
        issue_details_alias!(13),
        issue_details_alias!(14),
        issue_details_alias!(15),
        issue_details_alias!(16),
        issue_details_alias!(17),
        issue_details_alias!(18),
        issue_details_alias!(19),
        issue_details_alias!(20),
        issue_details_alias!(21),
        issue_details_alias!(22),
        issue_details_alias!(23),
        "\n    }",
        board_issue!()
    );

    /// One alias of [`ISSUE_COPY_DETAILS`]: the item a batch's `$id<n>` names, with the first
    /// page of its comments and the issues it is blocked by.
    macro_rules! issue_copy_details_alias {
        ($n:literal) => {
            concat!(
                "\n      i",
                stringify!($n),
                ":node(id:$id",
                stringify!($n),
                "){__typename ...BoardIssue ... on Issue{comments(first:$first){nodes{",
                issue_comment!(),
                "}pageInfo{hasNextPage endCursor}} blockedBy(first:$nestedFirst){nodes{...Related}pageInfo{hasNextPage endCursor}}}}"
            )
        };
    }

    /// [`ISSUE_DETAILS`] as a copy reads the tasks it carries: each item with the first page of
    /// its comments and its `blockedBy` — the edges a copy reads of every task next, which
    /// [`ISSUE`] carries and [`ISSUE_DETAILS`] does not — [`COPY_DETAIL_BATCH`](super::COPY_DETAIL_BATCH)
    /// at a time, so a copy of a changed task costs this read and nothing else.
    ///
    /// A document of its own rather than a widening of [`ISSUE_DETAILS`], whose batch size is
    /// what `task show-many` is priced by and which reads no edge.
    ///
    /// `children`, read only while `$children` holds, is which issues the issue `$project` has
    /// as sub-issues — each one's id and title, the page `$nestedFirst` allows — so that a
    /// project copy, which reads its project and every task at once, knows from this same read
    /// which tasks the project holds at the destination, and its orphan check is answered by it
    /// rather than by a listing of the project.
    pub const ISSUE_COPY_DETAILS: &str = concat!(
        r#"query($id0:ID!,$id1:ID!,$id2:ID!,$id3:ID!,$id4:ID!,$id5:ID!,$id6:ID!,$id7:ID!,$id8:ID!,$id9:ID!,$id10:ID!,$id11:ID!,$id12:ID!,$id13:ID!,$id14:ID!,$id15:ID!,$id16:ID!,$id17:ID!,$id18:ID!,$id19:ID!,$id20:ID!,$project:ID!,$children:Boolean!,$first:Int!,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){"#,
        issue_copy_details_alias!(0),
        issue_copy_details_alias!(1),
        issue_copy_details_alias!(2),
        issue_copy_details_alias!(3),
        issue_copy_details_alias!(4),
        issue_copy_details_alias!(5),
        issue_copy_details_alias!(6),
        issue_copy_details_alias!(7),
        issue_copy_details_alias!(8),
        issue_copy_details_alias!(9),
        issue_copy_details_alias!(10),
        issue_copy_details_alias!(11),
        issue_copy_details_alias!(12),
        issue_copy_details_alias!(13),
        issue_copy_details_alias!(14),
        issue_copy_details_alias!(15),
        issue_copy_details_alias!(16),
        issue_copy_details_alias!(17),
        issue_copy_details_alias!(18),
        issue_copy_details_alias!(19),
        issue_copy_details_alias!(20),
        "\n      children:node(id:$project) @include(if:$children){... on Issue{id subIssues(first:$nestedFirst){nodes{id title} pageInfo{hasNextPage}}}}",
        "\n    }",
        board_issue!(),
        related_issue!()
    );

    /// Which issue one comment is on, read before that comment is edited or removed.
    ///
    /// GitHub's comment mutations take the comment's id and nothing else, so without this a
    /// comment id given against the wrong task would change a comment on another issue.
    pub const COMMENT_ISSUE: &str =
        r#"query($id:ID!){node(id:$id){__typename ... on IssueComment{id body issue{id}}}}"#;
    /// Adds one comment to an issue, signed as the account the token belongs to.
    pub const ADD_COMMENT: &str = concat!(
        r#"mutation($input:AddCommentInput!){addComment(input:$input){subject{id} commentEdge{node{"#,
        issue_comment!(),
        r#"}}}}"#
    );
    /// Replaces the body of one issue comment.
    pub const UPDATE_COMMENT: &str = concat!(
        r#"mutation($input:UpdateIssueCommentInput!){updateIssueComment(input:$input){issueComment{"#,
        issue_comment!(),
        r#"}}}"#
    );
    /// Removes one issue comment. Its payload carries nothing about the comment it removed.
    pub const DELETE_COMMENT: &str = r#"mutation($input:DeleteIssueCommentInput!){deleteIssueComment(input:$input){clientMutationId}}"#;

    /// The variables of the search slots `<n>` of [`SEARCH_BATCH`]: its search, where it
    /// resumes, its page size and whether it is asked.
    macro_rules! search_batch_variables {
        ($($n:literal)+) => {
            concat!($(
                ",$search", stringify!($n), ":String!,$after", stringify!($n),
                ":String,$first", stringify!($n), ":Int!,$ask", stringify!($n), ":Boolean!",
            )+)
        };
    }
    /// The search slots `<n>` of [`SEARCH_BATCH`]: [`SEARCH_ISSUES`]'s one connection, aliased
    /// `s<n>`, sent only while `$ask<n>` holds.
    macro_rules! search_batch_slots {
        ($($n:literal)+) => {
            concat!($(
                "\n      s", stringify!($n), ":search(query:$search", stringify!($n),
                ",type:$type,first:$first", stringify!($n), ",after:$after", stringify!($n),
                ") @include(if:$ask", stringify!($n),
                "){pageInfo{hasNextPage endCursor} nodes{__typename ...BoardIssue}}",
            )+)
        };
    }
    /// The variables of the origin slots `<n>` of [`SEARCH_BATCH`]: the board's field filter
    /// naming the origin, where its items resume, their page size and whether it is asked.
    macro_rules! origin_batch_variables {
        ($($n:literal)+) => {
            concat!($(
                ",$filter", stringify!($n), ":String!,$itemsAfter", stringify!($n),
                ":String,$itemsFirst", stringify!($n), ":Int!,$origin", stringify!($n),
                ":Boolean!",
            )+)
        };
    }
    /// The origin slots `<n>` of [`SEARCH_BATCH`]: [`ORIGIN_LOOKUP`]'s board half, the board's
    /// own items through its field filter, aliased `o<n>`, sent only while `$origin<n>` holds.
    macro_rules! origin_batch_slots {
        ($($n:literal)+) => {
            concat!($(
                "\n      o", stringify!($n), ":repositoryOwner(login:$owner) @include(if:$origin",
                stringify!($n),
                "){... on ProjectV2Owner{projectV2(number:$number){items(first:$itemsFirst",
                stringify!($n), ",after:$itemsAfter", stringify!($n), ",query:$filter",
                stringify!($n), "){nodes{id updatedAt ", board_item_values!(),
                board_item_content!(), "} pageInfo{hasNextPage endCursor}}}}}",
            )+)
        };
    }

    /// Up to [`SEARCH_BATCH`](super::SEARCH_BATCH) board-scoped issue searches in one request,
    /// each [`SEARCH_ISSUES`]'s connection in an aliased `s<n>` field with its own query, page
    /// size and cursor, and beside them up to
    /// [`SEARCH_BATCH_ORIGINS`](super::SEARCH_BATCH_ORIGINS) origin lookups' board halves —
    /// [`ORIGIN_LOOKUP`]'s `items` filtered by the origin field — in aliased `o<n>` fields, each
    /// origin's issue search in a search slot of its own.
    ///
    /// What `task find` asks a board, every search a page at a time: one request a round,
    /// rather than one per search. A slot not asked is left out by `@include`, its page size
    /// bound to one; GitHub answers an error about one slot at that slot's path, and the other
    /// slots are answered beside it.
    pub const SEARCH_BATCH: &str = concat!(
        r#"query($owner:String!,$number:Int!,$type:SearchType!,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        search_batch_variables!(0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19),
        origin_batch_variables!(0 1 2 3 4),
        "){",
        search_batch_slots!(0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19),
        origin_batch_slots!(0 1 2 3 4),
        "\n    }",
        board_issue!()
    );

    /// The comment slots `<n>` of [`ADD_COMMENTS`]: [`ADD_COMMENT`]'s mutation, aliased
    /// `comment<n>`, sent only while `$send<n>` holds.
    macro_rules! comment_slots {
        ($($n:literal)+) => {
            concat!($(
                " comment", stringify!($n), ":addComment(input:$comment", stringify!($n),
                ") @include(if:$send", stringify!($n), "){subject{id} commentEdge{node{",
                issue_comment!(), "}}}",
            )+)
        };
    }

    /// Adds up to [`COMMENT_BATCH`](super::COMMENT_BATCH) comments in one request, each as
    /// [`ADD_COMMENT`] adds one, in order: what `task comment add-many` sends. GitHub runs each
    /// alias on its own and answers one it refused at its own path, beside the others.
    pub const ADD_COMMENTS: &str = concat!(
        "mutation(",
        wide_variables!("comment", "AddCommentInput", "send"; 0 1 2 3 4 5 6 7 8 9 10),
        "){",
        comment_slots!(0 1 2 3 4 5 6 7 8 9 10),
        "}"
    );

    /// Many tasks' targeted updates, each as [`UPDATE_FIELDS`] and [`UPDATE_ISSUE`] write one:
    /// up to [`TASK_FIELD_BATCH`](super::TASK_FIELD_BATCH) board field values in `field<n>`,
    /// [`MUTATION_BATCH`](super::MUTATION_BATCH) field clears in `clear<n>`, and as many
    /// `updateIssue`s in `update<n>`, each sent only while its own `send…<n>` holds.
    ///
    /// What `task update-many` sends, in two requests at most: every item's field values and
    /// clears first, then every issue update of an item whose fields all landed — the order one
    /// item's update keeps, so a refusal part way leaves an item's body as it stood.
    ///
    /// Each `update<n>` selects the issue it updated as a read of it does, so its response spells
    /// the change token the task holds after its batch's writes — the version `update-many`
    /// answers — with no read after them. Its page sizes are bound by `$nestedFirst` and
    /// `$boardItems`, as every read's are.
    pub const WRITE_TASKS: &str = concat!(
        "mutation(",
        wide_variables!("field", "UpdateProjectV2ItemFieldValueInput", "sendField"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18),
        wide_variables!("clear", "ClearProjectV2ItemFieldValueInput", "sendClear"; 0 1 2 3 4 5 6 7 8 9 10),
        wide_variables!("update", "UpdateIssueInput", "sendUpdate"; 0 1 2 3 4 5 6 7 8 9 10),
        r#"$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        "){",
        wide_slots!("field", "updateProjectV2ItemFieldValue", "projectV2Item{id}", "sendField"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18),
        wide_slots!("clear", "clearProjectV2ItemFieldValue", "projectV2Item{id}", "sendClear"; 0 1 2 3 4 5 6 7 8 9 10),
        wide_slots!("update", "updateIssue", "issue{id ...BoardIssue}", "sendUpdate"; 0 1 2 3 4 5 6 7 8 9 10),
        "}",
        board_issue!()
    );

    /// One alias of [`ISSUE_RELATIONS`]: the item a batch's `$id<n>` names, as a read of it by
    /// its own id reads it, with a page of its sub-issues' ids.
    macro_rules! issue_relations_alias {
        ($n:literal) => {
            concat!(
                "\n      i",
                stringify!($n),
                ":node(id:$id",
                stringify!($n),
                "){__typename ...BoardIssue ... on Issue{subIssues(first:$first){nodes{id}pageInfo{hasNextPage}}}}"
            )
        };
    }

    /// [`ISSUE_DETAILS`] with each issue's sub-issues in place of its comments: what `task
    /// relations` and a link write read of [`DETAIL_BATCH`](super::DETAIL_BATCH) tasks at once
    /// on a source reading sub-issues as children — each task whole, its parent and the ids of up
    /// to `$first` of its children.
    pub const ISSUE_RELATIONS: &str = concat!(
        r#"query($id0:ID!,$id1:ID!,$id2:ID!,$id3:ID!,$id4:ID!,$id5:ID!,$id6:ID!,$id7:ID!,$id8:ID!,$id9:ID!,$id10:ID!,$id11:ID!,$id12:ID!,$id13:ID!,$id14:ID!,$id15:ID!,$id16:ID!,$id17:ID!,$id18:ID!,$id19:ID!,$id20:ID!,$id21:ID!,$id22:ID!,$id23:ID!,$first:Int!,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
        relationship_events_variable!(),
        r#"){"#,
        issue_relations_alias!(0),
        issue_relations_alias!(1),
        issue_relations_alias!(2),
        issue_relations_alias!(3),
        issue_relations_alias!(4),
        issue_relations_alias!(5),
        issue_relations_alias!(6),
        issue_relations_alias!(7),
        issue_relations_alias!(8),
        issue_relations_alias!(9),
        issue_relations_alias!(10),
        issue_relations_alias!(11),
        issue_relations_alias!(12),
        issue_relations_alias!(13),
        issue_relations_alias!(14),
        issue_relations_alias!(15),
        issue_relations_alias!(16),
        issue_relations_alias!(17),
        issue_relations_alias!(18),
        issue_relations_alias!(19),
        issue_relations_alias!(20),
        issue_relations_alias!(21),
        issue_relations_alias!(22),
        issue_relations_alias!(23),
        "\n    }",
        board_issue!()
    );

    /// Up to [`RELATION_BATCH`](super::RELATION_BATCH) parent writes in one request:
    /// `parent<n>` an `addSubIssue` (a re-parent carrying `replaceParent: true`) and `unparent<n>`
    /// a `removeSubIssue`, each sent only while its own `send…<n>` holds. GitHub runs each alias
    /// on its own and answers one it refused — a cycle, a depth or children limit — at its own
    /// path, beside the others.
    pub const SET_PARENTS: &str = concat!(
        "mutation(",
        wide_variables!("parent", "AddSubIssueInput", "sendParent"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29),
        wide_variables!("unparent", "RemoveSubIssueInput", "sendUnparent"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29),
        "){",
        wide_slots!("parent", "addSubIssue", "issue{id} subIssue{id}", "sendParent"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29),
        wide_slots!("unparent", "removeSubIssue", "issue{id} subIssue{id}", "sendUnparent"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29),
        "}"
    );

    /// Up to [`RELATION_BATCH`](super::RELATION_BATCH) `relates-to` relations added in one
    /// request, each `relate<n>` sent only while `$send<n>` holds: what a link of a type a
    /// source mirrors natively writes beside its metadata.
    pub const ADD_RELATES_TO: &str = concat!(
        "mutation(",
        wide_variables!("relate", "AddRelatesToInput", "send"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29),
        "){",
        wide_slots!("relate", "addRelatesTo", "issue{id} relatedIssue{id}", "send"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29),
        "}"
    );

    /// [`ADD_RELATES_TO`]'s removal: up to [`RELATION_BATCH`](super::RELATION_BATCH)
    /// `relates-to` relations removed in one request.
    pub const REMOVE_RELATES_TO: &str = concat!(
        "mutation(",
        wide_variables!("unrelate", "RemoveRelatesToInput", "send"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29),
        "){",
        wide_slots!("unrelate", "removeRelatesTo", "issue{id} relatedIssue{id}", "send"; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29),
        "}"
    );

    /// One alias of [`RELATES_TO_READS`]: the issue `$id<n>` names and the ids of up to
    /// `$first` of the issues it relates to.
    macro_rules! relates_to_alias {
        ($n:literal) => {
            concat!(
                "\n      r",
                stringify!($n),
                ":node(id:$id",
                stringify!($n),
                "){... on Issue{id relatesTo(first:$first){nodes{id}pageInfo{hasNextPage}}}}"
            )
        };
    }

    /// What [`DETAIL_BATCH`](super::DETAIL_BATCH) issues relate to, one alias each: read
    /// after a relation write whose answer never arrived, so each relation is answered from what
    /// GitHub holds rather than sent again blindly.
    pub const RELATES_TO_READS: &str = concat!(
        r#"query($id0:ID!,$id1:ID!,$id2:ID!,$id3:ID!,$id4:ID!,$id5:ID!,$id6:ID!,$id7:ID!,$id8:ID!,$id9:ID!,$id10:ID!,$id11:ID!,$id12:ID!,$id13:ID!,$id14:ID!,$id15:ID!,$id16:ID!,$id17:ID!,$id18:ID!,$id19:ID!,$id20:ID!,$id21:ID!,$id22:ID!,$id23:ID!,$first:Int!){"#,
        relates_to_alias!(0),
        relates_to_alias!(1),
        relates_to_alias!(2),
        relates_to_alias!(3),
        relates_to_alias!(4),
        relates_to_alias!(5),
        relates_to_alias!(6),
        relates_to_alias!(7),
        relates_to_alias!(8),
        relates_to_alias!(9),
        relates_to_alias!(10),
        relates_to_alias!(11),
        relates_to_alias!(12),
        relates_to_alias!(13),
        relates_to_alias!(14),
        relates_to_alias!(15),
        relates_to_alias!(16),
        relates_to_alias!(17),
        relates_to_alias!(18),
        relates_to_alias!(19),
        relates_to_alias!(20),
        relates_to_alias!(21),
        relates_to_alias!(22),
        relates_to_alias!(23),
        "\n    }"
    );

    /// What a mutation of a write the store's cache expects selects of the issue it wrote:
    /// everything [`ISSUE_DETAIL`] reads of it, so its response shows the task whole — its
    /// change token's every part and the first page of its comments.
    ///
    /// Sent **only** while that write runs, in place of the document named beside each
    /// below, so every other write's documents, prices and request counts are what they
    /// were. GitHub's token waits for a task to settle (`change_token::SETTLED`), and a
    /// mutation's response is stamped seconds after the change it made, so against GitHub
    /// itself a response rarely determines a version; the cache is then left unserveable
    /// for the task, and its next read refreshes it.
    macro_rules! whole_issue {
        () => {
            concat!(
                r#"__typename ...BoardIssue ... on Issue{comments(first:$first){nodes{"#,
                issue_comment!(),
                r#"}pageInfo{hasNextPage endCursor}}}"#
            )
        };
    }

    /// The variables every document of `whole_issue!` declares beside its own.
    macro_rules! whole_variables {
        () => {
            concat!(
                r#",$first:Int!,$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
                relationship_events_variable!()
            )
        };
    }

    /// [`ADD_COMMENT`], selecting the whole issue commented on.
    pub const ADD_COMMENT_WHOLE: &str = concat!(
        r#"mutation($input:AddCommentInput!"#,
        whole_variables!(),
        r#"){addComment(input:$input){subject{id "#,
        whole_issue!(),
        r#"} commentEdge{node{"#,
        issue_comment!(),
        r#"}}}}"#,
        board_issue!()
    );
    /// [`UPDATE_COMMENT`], selecting the whole issue the comment is on.
    pub const UPDATE_COMMENT_WHOLE: &str = concat!(
        r#"mutation($input:UpdateIssueCommentInput!"#,
        whole_variables!(),
        r#"){updateIssueComment(input:$input){issueComment{"#,
        issue_comment!(),
        r#" issue{"#,
        whole_issue!(),
        r#"}}}}"#,
        board_issue!()
    );
    /// [`UPDATE_ISSUE`], selecting the whole issue updated.
    pub const UPDATE_ISSUE_WHOLE: &str = concat!(
        r#"mutation($input:UpdateIssueInput!"#,
        whole_variables!(),
        r#"){updateIssue(input:$input){issue{id "#,
        whole_issue!(),
        r#"}}}"#,
        board_issue!()
    );
    /// [`UPDATE_FIELD`], selecting the whole issue behind the board item written.
    pub const UPDATE_FIELD_WHOLE: &str = concat!(
        r#"mutation($input:UpdateProjectV2ItemFieldValueInput!,$readPriority:Boolean!,$priorityName:String!"#,
        whole_variables!(),
        r#"){updateProjectV2ItemFieldValue(input:$input){projectV2Item{id fieldValueByName(name:$priorityName) @include(if:$readPriority){... on ProjectV2ItemFieldSingleSelectValue{name field{... on ProjectV2SingleSelectField{id name options{id name}}}}} content{"#,
        whole_issue!(),
        r#"}}}}"#,
        board_issue!()
    );
    /// [`CLEAR_FIELD`], selecting the whole issue behind the board item cleared.
    pub const CLEAR_FIELD_WHOLE: &str = concat!(
        r#"mutation($input:ClearProjectV2ItemFieldValueInput!,$readPriority:Boolean!,$priorityName:String!"#,
        whole_variables!(),
        r#"){clearProjectV2ItemFieldValue(input:$input){projectV2Item{id fieldValueByName(name:$priorityName) @include(if:$readPriority){... on ProjectV2ItemFieldSingleSelectValue{name field{... on ProjectV2SingleSelectField{id name options{id name}}}}} content{"#,
        whole_issue!(),
        r#"}}}}"#,
        board_issue!()
    );

    /// The variables every document selecting [`board_issue_fields!`] inline declares beside
    /// its own.
    macro_rules! token_variables {
        () => {
            concat!(
                r#",$nestedFirst:Int!,$boardItems:Int!,$duplicates:Boolean!"#,
                relationship_events_variable!()
            )
        };
    }

    /// [`UPDATE_ISSUE`], selecting what a read of the issue updated selects — every part of its
    /// change token, and no comment — inline, so its root fields are the plain document's.
    ///
    /// What a targeted update sends in the plain document's place, so the version it answers
    /// is the one its own last response spells (`GitHubProjectsSource::written_tokens`). It
    /// costs the one point the plain document does.
    pub const UPDATE_ISSUE_TOKEN: &str = concat!(
        r#"mutation($input:UpdateIssueInput!"#,
        token_variables!(),
        r#"){updateIssue(input:$input){issue{"#,
        board_issue_fields!(),
        r#"}}}"#
    );
    /// [`UPDATE_FIELD`], selecting the issue behind the board item written as
    /// [`UPDATE_ISSUE_TOKEN`] selects the issue it updates.
    pub const UPDATE_FIELD_TOKEN: &str = concat!(
        r#"mutation($input:UpdateProjectV2ItemFieldValueInput!,$readPriority:Boolean!,$priorityName:String!"#,
        token_variables!(),
        r#"){updateProjectV2ItemFieldValue(input:$input){projectV2Item{id fieldValueByName(name:$priorityName) @include(if:$readPriority){... on ProjectV2ItemFieldSingleSelectValue{name field{... on ProjectV2SingleSelectField{id name options{id name}}}}} content{... on Issue{"#,
        board_issue_fields!(),
        r#"}}}}}"#
    );
    /// [`CLEAR_FIELD`], selecting the issue behind the board item cleared as
    /// [`UPDATE_ISSUE_TOKEN`] selects the issue it updates.
    pub const CLEAR_FIELD_TOKEN: &str = concat!(
        r#"mutation($input:ClearProjectV2ItemFieldValueInput!,$readPriority:Boolean!,$priorityName:String!"#,
        token_variables!(),
        r#"){clearProjectV2ItemFieldValue(input:$input){projectV2Item{id fieldValueByName(name:$priorityName) @include(if:$readPriority){... on ProjectV2ItemFieldSingleSelectValue{name field{... on ProjectV2SingleSelectField{id name options{id name}}}}} content{... on Issue{"#,
        board_issue_fields!(),
        r#"}}}}}"#
    );

    /// The document a targeted update sends in `document`'s place, when it has one: the twin
    /// whose answer spells the change token of the issue written.
    #[must_use]
    pub fn token_twin(document: &str) -> Option<&'static str> {
        [
            (UPDATE_ISSUE, UPDATE_ISSUE_TOKEN),
            (UPDATE_FIELD, UPDATE_FIELD_TOKEN),
            (CLEAR_FIELD, CLEAR_FIELD_TOKEN),
        ]
        .into_iter()
        .find(|(plain, _)| *plain == document)
        .map(|(_, token)| token)
    }

    /// The document sent in `document`'s place while a write the store's cache expects runs,
    /// when it has one.
    #[must_use]
    pub fn whole_twin(document: &str) -> Option<&'static str> {
        [
            (ADD_COMMENT, ADD_COMMENT_WHOLE),
            (UPDATE_COMMENT, UPDATE_COMMENT_WHOLE),
            (UPDATE_ISSUE, UPDATE_ISSUE_WHOLE),
            (UPDATE_FIELD, UPDATE_FIELD_WHOLE),
            (CLEAR_FIELD, CLEAR_FIELD_WHOLE),
        ]
        .into_iter()
        .find(|(plain, _)| *plain == document)
        .map(|(_, whole)| whole)
    }

    /// The node id of one person by their login, as the aliased field `u0` answering `$login0`:
    /// what an assignee write names a person by, since `assigneeIds` takes node ids and a person
    /// is named by login.
    ///
    /// The inventoried form of the lookup. What is sent is this document sized to the call
    /// ([`user_ids`]): one `u<n>:user(login:$login<n>)` field per person the call names that
    /// this process has not resolved, so one request answers every one of them. GitHub answers a
    /// login it does not know with `null` at that alias beside a `NOT_FOUND` error naming it,
    /// and the others' ids beside them.
    pub const USER_IDS: &str = "query($login0:String!){ u0:user(login:$login0){id login}}";

    /// [`USER_IDS`] for `people` logins — `$login0` to `$login<people - 1>`, each answered at
    /// its own alias — and [`USER_IDS`] itself for one.
    #[must_use]
    pub fn user_ids(people: usize) -> String {
        let people = people.max(1);
        let declared: Vec<String> = (0..people)
            .map(|slot| format!("$login{slot}:String!"))
            .collect();
        let fields: Vec<String> = (0..people)
            .map(|slot| format!("u{slot}:user(login:$login{slot}){{id login}}"))
            .collect();
        format!("query({}){{ {}}}", declared.join(","), fields.join(" "))
    }

    /// The document a request sending `document` with `variables` really sends: [`USER_IDS`]
    /// sized to the logins `variables` binds, and every other document as it is.
    #[must_use]
    pub fn sized<'a>(
        document: &'a str,
        variables: &serde_json::Value,
    ) -> std::borrow::Cow<'a, str> {
        if document != USER_IDS {
            return std::borrow::Cow::Borrowed(document);
        }
        let people = (0..)
            .take_while(|slot| variables.get(format!("login{slot}")).is_some())
            .count();
        std::borrow::Cow::Owned(user_ids(people))
    }

    /// The inventoried document `document` is an instance of: [`USER_IDS`] for any lookup
    /// [`user_ids`] sized, and `document` itself otherwise.
    #[must_use]
    pub fn inventoried(document: &str) -> &str {
        let people = document.matches(":user(login:$login").count();
        if people > 0 && user_ids(people) == document {
            USER_IDS
        } else {
            document
        }
    }

    /// The selection of an issue's — or a draft's — assignees, as a source opted in to them
    /// (`config.assignees: true`) adds it to every document that reads one; see
    /// [`with_assignees`].
    const ASSIGNEES: &str = " assignees(first:$assigneeFirst){nodes{id login}}";
    /// The variable [`ASSIGNEES`] pages by, declared with its value as its default — the most
    /// people GitHub assigns one issue — so no caller binds it, and inserted first among a
    /// document's variables.
    const ASSIGNEE_VARIABLE: &str = "$assigneeFirst:Int!=10 ";
    /// [`ASSIGNEE_VARIABLE`] for a document declaring no variables of its own.
    const ASSIGNEE_VARIABLES: &str = "($assigneeFirst:Int!=10)";
    /// Where [`ASSIGNEES`] goes: straight after each of these, wherever an issue or a draft is
    /// read — what every read of an issue selects of it — and inside the answer of each
    /// `createIssue` and `updateIssue`, so a write's own answer says whom GitHub assigned.
    const ASSIGNEE_MARKERS: [&str; 2] = [
        "commentCount:comments{totalCount}",
        "title body createdAt updatedAt",
    ];

    /// `document` as a source opted in to assignees sends it: [`ASSIGNEES`] selected wherever
    /// it reads an issue or a draft and in every `createIssue` and `updateIssue` answer, with the
    /// variable it pages by declared — or `None` when it reads and writes neither, and is sent
    /// as written.
    ///
    /// A rewrite of the one text sent rather than a second set of documents, so a source that
    /// has not opted in sends every document byte for byte as it did, and an opted-in one rides
    /// its assignees on the very reads and writes it already makes. [`without_assignees`] is its
    /// exact inverse, which the loopback board answers a document by.
    #[must_use]
    pub fn with_assignees(document: &str) -> Option<String> {
        let mut spelled = String::with_capacity(document.len() + 256);
        let mut inserted = false;
        let mut rest = document;
        while !rest.is_empty() {
            // The earliest place one is due, of every kind.
            let next = ASSIGNEE_MARKERS
                .iter()
                .filter_map(|marker| rest.find(marker).map(|at| (at + marker.len(), true)))
                .chain(["createIssue(", "updateIssue("].iter().filter_map(|write| {
                    let at = rest.find(write)?;
                    let open = at + rest[at..].find('{')?;
                    rest[open..]
                        .starts_with("{issue{id")
                        .then_some((open + "{issue{id".len(), true))
                        .or(Some((open, false)))
                }))
                .min_by_key(|(at, _)| *at);
            let Some((at, insert)) = next else {
                spelled.push_str(rest);
                break;
            };
            spelled.push_str(&rest[..at]);
            if insert {
                spelled.push_str(ASSIGNEES);
                inserted = true;
            }
            rest = &rest[at..];
            if !insert {
                // A write whose answer reads no issue's id first: past its opening brace.
                spelled.push('{');
                rest = &rest[1..];
            }
        }
        if !inserted {
            return None;
        }
        let open = spelled.find('(');
        let body = spelled.find('{');
        Some(match (open, body) {
            (Some(open), Some(body)) if open < body => {
                format!(
                    "{}{ASSIGNEE_VARIABLE}{}",
                    &spelled[..=open],
                    &spelled[open + 1..]
                )
            }
            (_, Some(body)) => format!(
                "{}{ASSIGNEE_VARIABLES}{}",
                &spelled[..body],
                &spelled[body..]
            ),
            _ => return None,
        })
    }

    /// `document` as [`with_assignees`] spelled it, back as the document it was written as — or
    /// `None` for a document it did not rewrite.
    #[must_use]
    pub fn without_assignees(document: &str) -> Option<String> {
        if !document.contains(ASSIGNEES) {
            return None;
        }
        let plain = document.replace(ASSIGNEES, "");
        Some(if plain.contains(ASSIGNEE_VARIABLES) {
            plain.replacen(ASSIGNEE_VARIABLES, "", 1)
        } else {
            plain.replacen(ASSIGNEE_VARIABLE, "", 1)
        })
    }

    /// Every document above, with what this source is doing when it sends one.
    ///
    /// One list rather than a `match` beside the constants: a rate-limit diagnostic has to
    /// name the call that was refused, and a `match` with a catch-all arm would answer a
    /// document added later with "talking to GitHub" and never say so.
    ///
    /// `documents_are_all_inventoried` reads this file back and fails naming any `pub
    /// const` here that this list omits, so the two cannot part — which is the same guard
    /// `CATEGORIES` carries, in the one shape available to a set of `&str` constants.
    pub const DOCUMENTS: [(&str, &str); 69] = [
        (USER_IDS, "looking up the people an assignee write names"),
        (
            ISSUE_RELATIONS,
            "reading several issues with their parents and sub-issues",
        ),
        (SET_PARENTS, "writing several issues' parents"),
        (ADD_RELATES_TO, "adding several relates-to relations"),
        (REMOVE_RELATES_TO, "removing several relates-to relations"),
        (
            RELATES_TO_READS,
            "reading what several issues relate to after a relation write went unanswered",
        ),
        (
            ISSUES_BY_CREATION,
            "listing a repository's issues newest first for a create key",
        ),
        (SEARCH_ISSUES, "searching this board's issues"),
        (ISSUE, "reading one issue"),
        (
            DOCUMENT_IN_PROJECT,
            "reading one document with its project and the project's members",
        ),
        (
            ISSUE_BOARD_ITEMS,
            "reading one issue's board memberships past the page it came with",
        ),
        (SUB_ISSUES, "reading a project's tasks"),
        (BOARD, "reading the board"),
        (
            BOARD_WITH_COMMENTS,
            "reading a board page with first comments",
        ),
        (ORIGIN_LOOKUP, "looking up the items copied from one origin"),
        (CLOSED_ITEMS, "reading the board's closed items"),
        (
            ISSUES_BY_REFERENCE,
            "reading a batch of issues by their references",
        ),
        (BOARD_FIELDS, "reading the board's fields"),
        (DRAFT, "reading one draft"),
        (REPOSITORY, "reading the destination repository"),
        (
            PROJECT_VISIBILITY,
            "reading whether the board's project is public",
        ),
        (
            CREATION_CONTEXT,
            "reading the board's fields and the destination repository",
        ),
        (ISSUE_DEPENDENCIES, "reading an issue's dependencies"),
        (ISSUE_DEPENDS_ON, "reading what an item depends on"),
        (ISSUE_DEPENDED_ON_BY, "reading what depends on an issue"),
        (CREATE_ISSUE, "creating an issue"),
        (
            ORIGIN_LOOKUPS,
            "looking up the items copied from several origins, and what creating one reads",
        ),
        (
            ORIGIN_LOOKUPS_WITH_MEMBERS,
            "looking up the items copied from several origins, with a project's members",
        ),
        (CREATE_ISSUES, "creating several issues together"),
        (
            CREATED_BOARD_ITEMS,
            "reading the board items of the issues a copy created",
        ),
        (
            WRITE_FIELD_VALUES,
            "writing board fields of the items a copy created",
        ),
        (
            WRITE_FIELDS_AND_EDGES,
            "writing board fields and dependencies of the items a copy created",
        ),
        (CLOSE_ISSUES, "closing the issues a copy created"),
        (UPDATE_ISSUES, "updating several issues together"),
        (ADD_BLOCKED_BYS, "recording several dependencies together"),
        (ADD_TO_BOARD, "adding an issue to the board"),
        (UPDATE_ISSUE, "updating an issue"),
        (UPDATE_DRAFT, "updating a draft item"),
        (UPDATE_FIELD, "writing a board field"),
        (UPDATE_FIELDS, "writing board fields together"),
        (CLEAR_FIELD, "clearing a board field"),
        (
            CREATE_FIELD,
            "creating a board single-select field with its options",
        ),
        (
            STATUS_OPTIONS_SNAPSHOT,
            "snapshotting board Status options and assignments",
        ),
        (
            STATUS_OPTIONS_UPDATE,
            "safely replacing the board Status option list",
        ),
        (ADD_SUB_ISSUE, "filing an issue under its project"),
        (REMOVE_SUB_ISSUE, "taking an issue out of its project"),
        (ADD_BLOCKED_BY, "recording a dependency"),
        (REMOVE_BLOCKED_BY, "removing a dependency"),
        (DELETE_ISSUE, "deleting an issue"),
        (DELETE_ISSUES, "deleting several issues together"),
        (ISSUE_COMMENTS, "reading a task's comments"),
        (ISSUE_DETAIL, "reading one issue with its comments"),
        (
            ISSUE_DETAILS,
            "reading a batch of issues with their comments",
        ),
        (
            ISSUE_COPY_DETAILS,
            "reading a batch of issues with their comments and dependencies for a copy",
        ),
        (COMMENT_ISSUE, "reading which issue a comment is on"),
        (ADD_COMMENT, "adding a comment"),
        (UPDATE_COMMENT, "editing a comment"),
        (DELETE_COMMENT, "deleting a comment"),
        (
            SEARCH_BATCH,
            "searching this board's issues several ways at once",
        ),
        (ADD_COMMENTS, "adding several comments together"),
        (WRITE_TASKS, "updating several tasks together"),
        (
            ADD_COMMENT_WHOLE,
            "adding a comment, answered with the whole issue",
        ),
        (
            UPDATE_COMMENT_WHOLE,
            "editing a comment, answered with the whole issue",
        ),
        (
            UPDATE_ISSUE_WHOLE,
            "updating an issue, answered with it whole",
        ),
        (
            UPDATE_FIELD_WHOLE,
            "writing a board field, answered with the whole issue",
        ),
        (
            CLEAR_FIELD_WHOLE,
            "clearing a board field, answered with the whole issue",
        ),
        (
            UPDATE_ISSUE_TOKEN,
            "updating an issue, answered with its version",
        ),
        (
            UPDATE_FIELD_TOKEN,
            "writing a board field, answered with the issue's version",
        ),
        (
            CLEAR_FIELD_TOKEN,
            "clearing a board field, answered with the issue's version",
        ),
    ];
}

/// One aliased mutation field of [`graphql::UPDATE_FIELDS`]: the key its answer comes back
/// under, the variable carrying its input, and the boolean variable that includes it.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub struct FieldSlot {
    /// The key of the answer's `data` this field's payload is under.
    pub alias: &'static str,
    /// The variable carrying this field's input.
    pub variable: &'static str,
    /// The boolean variable whose `true` sends this field.
    pub include: &'static str,
}

/// The value writes [`graphql::UPDATE_FIELDS`] carries, in the order GitHub runs them.
pub const FIELD_WRITE_SLOTS: [FieldSlot; 6] = [
    FieldSlot {
        alias: "updateProjectV2ItemFieldValue",
        variable: "input",
        include: "writeFirst",
    },
    FieldSlot {
        alias: "second",
        variable: "second",
        include: "writeSecond",
    },
    FieldSlot {
        alias: "third",
        variable: "third",
        include: "writeThird",
    },
    FieldSlot {
        alias: "fourth",
        variable: "fourth",
        include: "writeFourth",
    },
    FieldSlot {
        alias: "fifth",
        variable: "fifth",
        include: "writeFifth",
    },
    FieldSlot {
        alias: "sixth",
        variable: "sixth",
        include: "writeSixth",
    },
];

/// The clears [`graphql::UPDATE_FIELDS`] carries, run after every write of the same request.
pub const FIELD_CLEAR_SLOTS: [FieldSlot; 3] = [
    FieldSlot {
        alias: "cleared",
        variable: "clear",
        include: "writeClear",
    },
    FieldSlot {
        alias: "clearedSecond",
        variable: "clearSecond",
        include: "writeClearSecond",
    },
    FieldSlot {
        alias: "clearedThird",
        variable: "clearThird",
        include: "writeClearThird",
    },
];

/// Which of GitHub's two rate limiters refused a request.
///
/// Waiting is the whole answer to the primary budget, and polling is what *extends* the
/// secondary one — so an operator told the wrong one takes the wrong next step, which is
/// the whole reason this is carried rather than collapsed into "rate limited".
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum Limiter {
    /// The hourly API budget, which `gh api rate_limit` reports and a wait answers.
    Primary,
    /// The burst limiter over content-generating requests, which nothing reports.
    Secondary,
}

/// The wordings GitHub answers a secondary rate limit with.
///
/// It sends them under a forbidden status, under a too-many-requests status, and inside
/// the `errors` of a *successful* response, which is why the text is what this matches on
/// rather than the status. `abuse detection` is the wording GitHub used before the
/// limiter was renamed and still returns from some endpoints; `submitted too quickly` is
/// what a burst of content creation is refused with.
///
/// This is GitHub's vocabulary rather than this source's, so it is pinned rather than
/// remembered: `tests/fixtures/rate-limits.json` records where each wording was read and
/// when, and the drift gate reconciles the two lists both ways. Public for that gate
/// alone — a caller has no use for it, and matching on a refusal is this source's job.
pub const SECONDARY_WORDINGS: [&str; 5] = [
    "secondary rate limit",
    "temporarily blocked from content creation",
    "abuse detection",
    "submitted too quickly",
    "exceeded a secondary",
];

/// The wordings GitHub answers an exhausted primary budget with.
///
/// `rate_limited` is the `type` its GraphQL error carries, which is read as a field rather
/// than looked for in the response text. `api rate limit already exceeded` is what GraphQL
/// answers a request made once the hour's budget is spent — "API rate limit already exceeded
/// for user ID …" in the `errors` of an HTTP 200, with no `type` — and neither of the other
/// two phrases is a substring of it, so without it that answer read as a refusal that will
/// never lift. Pinned and gated exactly as [`SECONDARY_WORDINGS`] is, and public for the same
/// one reason.
pub const PRIMARY_WORDINGS: [&str; 4] = [
    "api rate limit exceeded",
    "api rate limit already exceeded",
    "rate limit exceeded",
    "rate_limited",
];

/// The wordings GitHub answers a write that collided with another change to the same
/// ordering with, in the `errors` of an HTTP 200: "Your attempt to move this item created a
/// temporary conflict. Please try again." The write did not take effect and GitHub asks for it
/// again, so it is retried on the same growing schedule a rate limit without a hint is, within
/// the same retry budget — except a `createIssue`, whose answer cannot say the issue was not
/// made, and which is therefore reported rather than replayed into a duplicate.
///
/// Observed rather than documented, so it is pinned exactly as [`SECONDARY_WORDINGS`] is:
/// `tests/fixtures/rate-limits.json` records where it was read under `conflict.observed`, and
/// the same drift gate reconciles the two both ways. Public for that gate alone.
pub const CONFLICT_WORDINGS: [&str; 1] = ["created a temporary conflict"];

/// What a response *says about itself*, which is the only place a refusal can be read.
///
/// Deliberately not the whole response body. A board is a place people write about their
/// own work, and a task on it titled "the secondary rate limit" would, matched across the
/// raw text, turn a perfectly good answer into a refusal this source then waited out and
/// reported. So the item data is never read: what is read is GitHub's own REST-style
/// `message` envelope, which is what a forbidden status carries, and the `message` and
/// `type` of each GraphQL error, which is where a *successful* response says it.
///
/// A body that is not JSON at all has nothing structured to read, so only a failing
/// response's own text is taken — a successful response that is not JSON is malformed
/// rather than refused, and [`GitHubProjectsSource::answer`] says so.
fn refusal_wording(status: StatusCode, body: &str) -> String {
    let Ok(parsed) = serde_json::from_str::<Value>(body) else {
        return if status.is_success() {
            String::new()
        } else {
            body.to_owned()
        };
    };
    let mut said: Vec<&str> = parsed
        .get("message")
        .and_then(Value::as_str)
        .into_iter()
        .collect();
    if let Some(errors) = parsed.get("errors").and_then(Value::as_array) {
        for error in errors {
            said.extend(
                ["message", "type"]
                    .into_iter()
                    .filter_map(|key| error.get(key).and_then(Value::as_str)),
            );
        }
    }
    said.join("; ")
}

impl Limiter {
    /// Which limiter refused this response, or `None` when none of them did.
    ///
    /// The wording is read first and the status only decides what carries none of it,
    /// because GitHub answers a secondary limit with a forbidden status far more often
    /// than with too-many-requests — while a forbidden status saying nothing about a limit
    /// really is a credential this token lacks.
    ///
    /// A response is a refusal because of its status or its own wording. A spent budget
    /// only ever explains one; it never turns an answer into a refusal.
    fn classify(status: StatusCode, budget_exhausted: bool, body: &str) -> Option<Self> {
        let normalized = refusal_wording(status, body).to_ascii_lowercase();
        if SECONDARY_WORDINGS
            .iter()
            .any(|wording| normalized.contains(wording))
        {
            return Some(Self::Secondary);
        }
        if status == StatusCode::TOO_MANY_REQUESTS {
            return Some(Self::Primary);
        }
        // An exhausted budget *explains* a response that failed; it does not make one that
        // succeeded into a failure. GitHub sets `x-ratelimit-remaining: 0` on the last
        // request the budget allowed as well as on the ones it then refuses, so reading
        // the header alone threw away a good answer — and, once refusals were retried,
        // replayed a request that had already taken effect.
        if !status.is_success() && budget_exhausted {
            return Some(Self::Primary);
        }
        // A successful response saying it: GitHub reports a GraphQL rate limit in the
        // `errors` of an HTTP 200, where nothing about the status says so at all.
        if status.is_success()
            && PRIMARY_WORDINGS
                .iter()
                .any(|wording| normalized.contains(wording))
        {
            return Some(Self::Primary);
        }
        None
    }

    /// What this limiter is called where an operator can look it up.
    const fn name(self) -> &'static str {
        match self {
            Self::Primary => "GitHub's primary API rate limit",
            Self::Secondary => "GitHub's secondary rate limit",
        }
    }

    /// What the endpoint an operator would go and check says about this limiter.
    const fn where_to_look(self) -> &'static str {
        match self {
            Self::Primary => {
                "That is the budget `gh api rate_limit` reports, so that endpoint says when it \
                 comes back."
            }
            Self::Secondary => {
                "That limiter is not the primary API budget: `gh api rate_limit` reports the \
                 primary budget and does not report this one, so budget showing there says \
                 nothing about this refusal, and every further attempt extends it."
            }
        }
    }

    /// The next step this limiter actually calls for.
    const fn what_to_do(self) -> &'static str {
        match self {
            Self::Primary => {
                "wait for the reset `gh api rate_limit` reports, then run the command again."
            }
            Self::Secondary => {
                "leave this board alone for a few minutes, then run the command again — or \
                 raise pacing.min_mutation_interval_ms on this source so it writes more slowly."
            }
        }
    }
}

/// One rate-limit refusal, and the wait GitHub asked for if it asked for one.
#[derive(Debug, Clone, Copy)]
struct Limited {
    limiter: Limiter,
    hint: Option<u64>,
    /// Which header the hint came from: `retry-after`, or an exhausted primary budget's
    /// `x-ratelimit-reset`.
    hinted_by: Hint,
}

/// Where a refusal's hint came from, which is what its wait is named after.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum Hint {
    /// `retry-after`, or no hint at all.
    RetryAfter,
    /// An exhausted primary budget's `x-ratelimit-reset`.
    PrimaryReset,
}

impl Limited {
    /// The wait this refusal is answered with, when the backing-off schedule is at `backoff`,
    /// and the cause it is named by: the hint, raised to the schedule's own next wait —
    /// GitHub really does send `retry-after: 0`, and retrying at once is the one move that
    /// extends a secondary limit — or the schedule alone when there is none.
    fn wait(self, backoff: Duration) -> (Duration, WaitCause) {
        match (self.hint, self.hinted_by) {
            (Some(hint), Hint::RetryAfter) => (
                Duration::from_secs(hint).max(backoff),
                pacing::cause::RETRY_AFTER,
            ),
            (Some(hint), Hint::PrimaryReset) => (
                Duration::from_secs(hint).max(backoff),
                pacing::cause::PRIMARY_RESET,
            ),
            (None, _) => (backoff, pacing::cause::BACKOFF),
        }
    }

    /// What the caller is told once this source has waited as long as it may.
    ///
    /// Both limiters report as [`SourceError::RateLimited`], because that is what
    /// happened: the kind a caller matches on says a rate limit refused this, and nothing
    /// about *which* limiter it was makes it a different kind of failure. What differs is
    /// the operator's next step, and that is what the message carries — a secondary
    /// refusal read as a primary one sends an operator to `gh api rate_limit`, where the
    /// budget looks fine, and then back to retry the very burst that was refused.
    fn exhausted(
        self,
        doing: &str,
        waits: u32,
        waited: Duration,
        needed: Duration,
        budget: Duration,
    ) -> SourceError {
        SourceError::RateLimited {
            retry_after_seconds: self.hint,
            message: Some(format!(
                "{} refused this source while {doing}; it waited {} out over {} and was refused \
                 again, and the next wait of {} would take it past the {} one call may spend \
                 waiting. {} next: {}",
                self.limiter.name(),
                plural(waits, "refusal"),
                seconds(waited),
                seconds(needed),
                seconds(budget),
                self.limiter.where_to_look(),
                self.limiter.what_to_do(),
            )),
        }
    }
}

/// One HTTP attempt's result, with what its response said about the rate limit.
///
/// The two travel together so the record and the outcome are written from the same place:
/// what a response said about the budget is only readable while that response is in hand,
/// and what the attempt *meant* is only decidable once its body has been read.
struct Attempted {
    result: Result<Value, Attempt>,
    limits: accounting::RateLimit,
    /// GitHub's own reported cost for this call, for a document that asked for it.
    reported_cost: Option<u64>,
    response_outcome: accounting::ResponseOutcome,
}

/// One attempt's outcome: an error to report, or a rate limit to wait out.
enum Attempt {
    Failed(SourceError),
    Limited(Limited),
    /// GitHub refused a write as a temporary conflict and asked for it again — see
    /// [`CONFLICT_WORDINGS`]. Carries what GitHub said, for the diagnostic once retrying ends.
    Conflicted(String),
}

fn plural(count: u32, thing: &str) -> String {
    if count == 1 {
        format!("{count} {thing}")
    } else {
        format!("{count} {thing}s")
    }
}

fn seconds(duration: Duration) -> String {
    format!("{:.1}s", duration.as_secs_f64())
}

/// The member [`stamp_served`] writes onto every node of a response that says when it last
/// changed: the `Date` that response was served at. No GitHub type has a field of this name.
const SERVED_AT: &str = "onetaskgraphServedAt";

/// Write `served` onto every object of `data` carrying an `updatedAt`, so whatever reads a
/// node's time later knows the server's clock on the response that read it — which is what
/// decides whether that node's change token is given (see [`change_token::SETTLED`]).
fn stamp_served(data: &mut Value, served: &str) {
    match data {
        Value::Object(members) => {
            if members.contains_key("updatedAt") {
                members.insert(SERVED_AT.to_owned(), Value::String(served.to_owned()));
            }
            for member in members.values_mut() {
                stamp_served(member, served);
            }
        }
        Value::Array(items) => {
            for item in items {
                stamp_served(item, served);
            }
        }
        _ => {}
    }
}

/// A header GitHub spells as a whole number of seconds, or `None` when this one is not.
///
/// A value that is present and unreadable is deliberately *not* an error. `retry-after` is
/// allowed by HTTP to be a date rather than a count, an intermediary can rewrite either
/// header, and neither is what makes a response a refusal — so the whole cost of one this
/// cannot read is that the refusal carries no hint and the backing-off schedule answers it
/// instead. Refusing the response over the header would turn a readable refusal into an
/// unreadable one, and refusing to *wait* would be the one wrong direction to fail in.
fn whole_seconds(value: Option<&reqwest::header::HeaderValue>) -> Option<u64> {
    value
        .and_then(|value| value.to_str().ok())
        .and_then(|value| value.trim().parse::<u64>().ok())
}

/// A transport failure with every cause beneath it, outermost first.
///
/// The client's own text for a request it could not complete is `error sending request for
/// url (…)`, and the socket error that says why — a refused or reset connection, an exhausted
/// port range, a connection closed before its answer — is only in the chain of sources below
/// it. Those differ in what to do next, so all of them are kept.
pub(crate) fn transport_error(error: &(dyn std::error::Error + 'static)) -> String {
    let mut text = error.to_string();
    let mut cause = error.source();
    while let Some(next) = cause {
        let said = next.to_string();
        if !text.ends_with(&said) {
            text = format!("{text}: {said}");
        }
        cause = next.source();
    }
    text
}

/// Every mutation this source sends creates content — an issue, a board item, a field of
/// one, a sub-issue link, a dependency, a comment — or edits or removes content of that
/// kind, and no query in [`graphql::DOCUMENTS`] does, so what the secondary limiter counts
/// and what the keyword says are the same set. That is what makes the keyword a sound test
/// rather than a convenient one: counting an edit or a removal the limiter might not have
/// counted costs a wait, and not counting one it did count costs the next fifty minutes.
pub(crate) fn is_mutation(query: &str) -> bool {
    query.trim_start().starts_with("mutation")
}

/// What this source was doing, for a diagnostic that has to say so.
///
/// Read out of [`graphql::DOCUMENTS`], which is the inventory rather than a copy of it, so
/// a document added without a description is caught by that list's own gate instead of
/// falling through to the vague arm below.
fn operation_description(query: &str) -> &'static str {
    let query = graphql::inventoried(query);
    graphql::DOCUMENTS
        .iter()
        .find(|(document, _)| *document == query)
        .map_or("talking to GitHub", |(_, doing)| *doing)
}

/// GitHub's published ceiling on content-generating requests in any one minute.
///
/// Pinned in `tests/fixtures/rate-limits.json` and gated against it, because it is
/// GitHub's number rather than this source's. This source paces at it over its window: a
/// content-creating request goes out at once while the minute's creations leave room for it,
/// and waits until they do otherwise — see the `pacing` module.
pub const CONTENT_CREATION_PER_MINUTE: u64 = 80;
/// The same ceiling as GitHub publishes it for any one hour, paced over its window as the
/// per-minute one is. Pinned beside its sibling so the gate would notice either one moving.
pub const CONTENT_CREATION_PER_HOUR: u64 = 500;
/// GitHub's published ceiling on secondary points in any one minute: what every GraphQL
/// request is charged against, a mutation [`SECONDARY_POINTS_PER_MUTATION`] and a query
/// [`SECONDARY_POINTS_PER_QUERY`]. Pinned and paced as [`CONTENT_CREATION_PER_MINUTE`] is.
pub const SECONDARY_POINTS_PER_MINUTE: u64 = 2_000;
/// The secondary points one GraphQL request carrying a mutation is charged.
pub const SECONDARY_POINTS_PER_MUTATION: u64 = 5;
/// The secondary points one GraphQL request carrying no mutation is charged.
pub const SECONDARY_POINTS_PER_QUERY: u64 = 1;
/// The fastest *steady* spacing of content-creating mutations that cannot exceed
/// [`CONTENT_CREATION_PER_MINUTE`], in milliseconds: 60000/80 is 750.
///
/// It is no longer the default. This source paces by GitHub's windows instead, so a copy's
/// few dozen writes go out as fast as GitHub answers them while the minute has room, rather
/// than 750 ms apart — which was most of a plan copy's wall time. An installation that wants
/// mutations spaced as well says so through `pacing.min_mutation_interval_ms`, and this is the
/// value that holds a long sustained burst under the per-minute bound by spacing alone.
pub const MIN_MUTATION_INTERVAL_MS: u64 = 60_000 / CONTENT_CREATION_PER_MINUTE;
/// First wait when a rate-limit refusal carries no hint; each further wait doubles it.
///
/// A doubling schedule from one second reaches a minute in six waits, which is GitHub's
/// own advice for a secondary limit — wait, and wait longer each time — without spending
/// the first minute of a transient refusal doing nothing.
pub const RETRY_BACKOFF_MS: u64 = 1_000;
/// Total time one call may spend waiting out rate limits before it reports a failure.
///
/// Two minutes is long enough to ride out the refusals a paced copy still collects and
/// short enough that a command an operator is watching returns. The bound is what makes
/// the wait a wait rather than a hang: a call refused past it ends in a diagnostic naming
/// the limiter, not in a process nobody can tell from a wedged one.
pub const RETRY_BUDGET_MS: u64 = 120_000;

fn default_token_env() -> String {
    "GH_PROJECTS_TOKEN".to_owned()
}
fn default_endpoint() -> String {
    "https://api.github.com/graphql".to_owned()
}

/// The name of a `Status` single-select option on the board.
///
/// Validated on the way in rather than checked later, so a blank option name — which
/// nothing on a board can be — is a state this type cannot hold.
#[derive(Debug, Clone, PartialEq, Eq, Deserialize, Serialize, schemars::JsonSchema)]
#[serde(try_from = "String")]
#[schemars(extend("minLength" = 1))]
pub struct ColumnName(String);

impl ColumnName {
    /// The option name, as the board spells it.
    fn as_str(&self) -> &str {
        &self.0
    }
}

impl TryFrom<String> for ColumnName {
    type Error = String;

    fn try_from(name: String) -> Result<Self, Self::Error> {
        if name.trim().is_empty() {
            return Err("a status_mapping option name cannot be blank".to_owned());
        }
        Ok(Self(name))
    }
}

/// The two closed states this product can mean.
///
/// GitHub's `IssueClosedStateReason` also spells `DUPLICATE`, which nothing here ever writes:
/// an issue closed as one *reads* `cancelled`, because its work is another issue's, but this
/// product never decides on a person's behalf that two issues are the same work.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize, schemars::JsonSchema)]
#[serde(rename_all = "kebab-case")]
pub enum ClosedState {
    /// `COMPLETED` — precisely done.
    Completed,
    /// `NOT_PLANNED` — precisely cancelled.
    NotPlanned,
}

impl ClosedState {
    const fn reason(self) -> &'static str {
        match self {
            Self::Completed => "COMPLETED",
            Self::NotPlanned => "NOT_PLANNED",
        }
    }
}

/// Configuration for one GitHub Projects v2 board.
///
/// It serializes to a document this same type reads back as itself, so a caller writing a
/// configuration — a consumer filing onto a board it configures — can build one here rather
/// than spell the shape again; an empty `metadata_fields` is left out, as one never set.
#[derive(Debug, Clone, Default, PartialEq, Eq, Deserialize, Serialize, schemars::JsonSchema)]
#[serde(default, deny_unknown_fields)]
pub struct GitHubProjectsConfig {
    /// Login of the user or organization which owns the board.
    pub owner: String, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `new` validates GitHub's owner grammar before private construction.
    /// The project number shown in the board's GitHub URL.
    pub project_number: u32, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `new` bounds this to a positive GraphQL Int.
    // llmlint: ignore-block[contracts_have_one_source_or_a_drift_gate] This doc is the field's schema description, which is what a person configuring the source reads, so it has to say when the field decides an issue's repository and when the item's own field does; the rule's one executable source is `GitHubProjectsSource::creation_target`, and `tests/plugin.rs` drives each case named here against the loopback board.
    /// `owner/name` of the repository this source creates an issue in when the item's own
    /// `repositories` field does not decide it.
    ///
    /// An item naming exactly one repository is created there — on a source declared
    /// `visibility: private`, only when that repository reads private, and otherwise as an
    /// item naming none is; a task or a document naming none or several is created in its
    /// parent project's repository; and a project, or a task or document with no parent,
    /// naming none or several is created here. A board
    /// has no repository of its own and `createIssue` requires one, so a write without
    /// this is refused naming the field. Reads never need it.
    pub repository: Option<String>, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `new` validates the `owner/name` grammar before private construction.
    // llmlint: ignore-end[contracts_have_one_source_or_a_drift_gate]
    /// GitHub logins whose repositories this source may create an issue in.
    ///
    /// Absent or empty, the default, restricts nothing. Otherwise a create whose issue the
    /// rule `repository` describes would file in a repository owned by anyone else is refused
    /// before anything of it is written — by `task copy`, `task copy --create`, a member
    /// create of a project copy and a dry run of each — with the error kind `creation-owner`,
    /// naming the repository and these owners. Logins compare case-insensitively, as GitHub
    /// compares them. An update of an issue that already exists is never refused by it.
    /// `task locate` reports such a refusal as `creates_in.refused` rather than failing.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    // Kept in the schema as `"default": []` although a serialized configuration leaves an empty
    // list out, so both SDKs model an absent list as an empty one rather than as `null`.
    #[schemars(!skip_serializing_if)]
    pub creation_owners: Vec<String>, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `new` validates each entry against GitHub's login grammar, naming the one it refuses.
    /// Environment variable containing a fine-grained token with Projects and Issues
    /// read/write plus Pull requests read-only access for every repository represented on
    /// the board.
    #[serde(default = "default_token_env")]
    pub token_env: String, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `new` validates the environment-variable grammar.
    /// GraphQL endpoint. GitHub Enterprise installations may override it.
    #[serde(default = "default_endpoint")]
    pub endpoint: String, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `new` converts it to the private validated `Url`.
    /// Per-instance mapping from a status category to the option of the board's one
    /// `Status` field it lands on, for a task and for a project.
    ///
    /// The shared `StatusMapping` grammar: each value is one option name for both kinds,
    /// `null` to disable the category for both, or `{task, project}` naming it per kind,
    /// where a kind left out leaves the category unmapped for that kind. A category this
    /// does not mention keeps its shipped default for both kinds: `backlog` to "Backlog",
    /// `todo` to "Todo", `queued` to "Queued", `in-progress` to "In Progress", `done` to
    /// "Done" plus closed as completed, `cancelled` to "Cancelled" plus closed as not
    /// planned, and `draft` and `unknown` unmapped. A category it does mention gets no
    /// shipped default for a kind it leaves out. `done` and `cancelled` close the issue for
    /// either kind. No two categories may name one option for the same kind, ignoring case.
    /// `unknown` may name one existing option; every unknown word then lands on it and
    /// reads back as `unknown` under its name. Unlike `local-md`, this source cannot keep
    /// each unknown word because it never creates board options.
    #[serde(default)]
    pub status_mapping: StatusMapping,
    /// Per-instance mapping from a task's priority to an option of this board's
    /// single-select field named `Priority`.
    ///
    /// Absent, this source holds no priority: every task reads as `none`, and a write of any
    /// other priority is refused before it reaches this board. Present, each of `urgent`,
    /// `high`, `medium` and `low` it does not mention keeps its shipped default — `Urgent`,
    /// `High`, `Medium` and `Low` — and an item with no value in the `Priority` field reads
    /// as `none`, so writing `none` clears the value. Option names match case-insensitively;
    /// no two levels may name one option. Reads and writes never create the field or an
    /// option: `onetaskgraph sources fields <source> --apply` does, and a write naming one
    /// the board lacks is refused pointing there.
    #[serde(default)]
    pub priority_mapping: Option<PriorityMappingConfig>,
    /// Caller metadata values this source also writes to board text fields, so the board can
    /// filter on them.
    ///
    /// Absent or empty, nothing is projected. Each entry names a board text `field`, a
    /// top-level metadata `key` and an optional `path` of object keys walked inside that key's
    /// value. On every item write — create, update and copy, of a task, a project or a
    /// document — a string found there is written to the field, and nothing is sent when the
    /// field already holds it; nothing there, or `null`, clears the field when it holds a
    /// value; any other JSON type is refused before any mutation. The metadata comment in the
    /// issue body stays the one home of the value, and every read reports metadata from it.
    /// Reads and writes never create the field: `onetaskgraph sources fields <source>
    /// --apply` does, as a text field, and a write to a board lacking it, or holding a
    /// non-text field of that name, is refused pointing there.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    // Kept in the schema as `"default": []` although a serialized configuration leaves an empty
    // list out, so both SDKs model an absent list as an empty one rather than as `null`.
    #[schemars(!skip_serializing_if)]
    pub metadata_fields: Vec<MetadataFieldConfig>,
    /// How fast this source writes, and how long it waits out a rate-limit refusal.
    ///
    /// Every field keeps its shipped default when it is absent, and the defaults are
    /// GitHub's own published limits rather than taste. See [`Pacing`].
    #[serde(default)]
    pub pacing: PacingConfig,
    /// What an issue's sub-issues mean on this board.
    ///
    /// `project`, the default, is this source's meaning since it was first written: an issue
    /// with sub-issues reads as a project and its sub-issues as that project's tasks. `parent`
    /// reads every issue as a task and a sub-issue as a child under the parent relation, which
    /// `task parent set-many` writes and `task relations` reads; a project write to such a
    /// source is refused naming this setting. The two meanings answer differently for the same
    /// issues, so a `parent` source needs a board of its own: two sources naming one board with
    /// different `sub_issues` are refused when the configuration is read.
    #[serde(default, skip_serializing_if = "SubIssues::is_project")]
    pub sub_issues: SubIssues,
    /// The link types this source also writes as GitHub's own `relates-to` relation, beside the
    /// store's metadata, so they show in the issue's Relationships panel: each type name mapped
    /// to `{"native": "relates-to"}`. A type not named here is kept in metadata alone. Any other
    /// word — `blocks`, `blocked-by`, `duplicate`, `similar` among them — is refused when the
    /// source is built.
    #[serde(default, skip_serializing_if = "BTreeMap::is_empty")]
    // Kept in the schema as `"default": {}` although a serialized configuration leaves an empty
    // map out, so both SDKs model an absent map as an empty one rather than as `null`.
    #[schemars(!skip_serializing_if)]
    pub link_types: BTreeMap<String, onetaskgraph_plugin_api::LinkTypeConfig>,
    /// Whether this source's tasks carry the GitHub users each issue is assigned to, and
    /// take an assignee write. `false`, the default, reads no issue's assignees and writes
    /// none: every request this source sends is then exactly what it was before assignees
    /// existed. `true` reads each issue's assignees on the reads it already makes, and writes
    /// one person or none — by login — in the `createIssue` or `updateIssue` the write already
    /// sends, refusing a person GitHub does not know or leaves off the issue.
    #[serde(default, skip_serializing_if = "std::ops::Not::not")]
    // Kept in the schema as `"default": false` although a serialized configuration leaves a
    // `false` out, so both SDKs model an absent setting as `false` rather than as `null`.
    #[schemars(!skip_serializing_if)]
    pub assignees: bool,
}

/// What an issue's sub-issues mean on a GitHub Projects board: see
/// [`GitHubProjectsConfig::sub_issues`].
#[derive(
    Debug, Clone, Copy, Default, PartialEq, Eq, Deserialize, Serialize, schemars::JsonSchema,
)]
#[serde(rename_all = "kebab-case")]
pub enum SubIssues {
    /// An issue with sub-issues is a project, and its sub-issues are its tasks.
    #[default]
    Project,
    /// Every issue is a task, and a sub-issue is a child under the parent relation.
    Parent,
}

impl SubIssues {
    /// Whether this is the default meaning, which a serialized configuration leaves out.
    #[must_use]
    pub const fn is_project(&self) -> bool {
        matches!(self, Self::Project)
    }
}

/// Which option of the board's `Priority` field each priority lands on.
///
/// One member per level rather than a map, so a key that is not a level is refused where
/// the configuration is read, naming the levels there are. `none` is not a member: it is no
/// value in the field, not an option of it.
#[derive(Debug, Clone, Default, PartialEq, Eq, Deserialize, Serialize, schemars::JsonSchema)]
#[serde(default, deny_unknown_fields)]
pub struct PriorityMappingConfig {
    /// The option `urgent` lands on; `Urgent` when absent.
    pub urgent: Option<PriorityOptionName>,
    /// The option `high` lands on; `High` when absent.
    pub high: Option<PriorityOptionName>,
    /// The option `medium` lands on; `Medium` when absent.
    pub medium: Option<PriorityOptionName>,
    /// The option `low` lands on; `Low` when absent.
    pub low: Option<PriorityOptionName>,
}

/// The name of an option of the board's `Priority` single-select field.
///
/// Validated on the way in, for the reason [`ColumnName`] is: nothing on a board can have a
/// blank name.
#[derive(Debug, Clone, PartialEq, Eq, Deserialize, Serialize, schemars::JsonSchema)]
#[serde(try_from = "String")]
#[schemars(extend("minLength" = 1))]
pub struct PriorityOptionName(String);

impl PriorityOptionName {
    /// The option name, as the board spells it.
    fn as_str(&self) -> &str {
        &self.0
    }
}

impl TryFrom<String> for PriorityOptionName {
    type Error = String;

    fn try_from(name: String) -> Result<Self, Self::Error> {
        if name.trim().is_empty() {
            return Err("a priority_mapping option name cannot be blank".to_owned());
        }
        Ok(Self(name))
    }
}

/// The name of the board field a priority is held in.
pub const PRIORITY_FIELD: &str = "Priority";

/// The four priorities a board option can hold, in the order a new `Priority` field lists
/// them. `none` is not among them: it is the field holding no value.
///
/// This list mirrors `Priority`, so it carries its own drift gate, in the shape [`CATEGORIES`]
/// does: [`level_position`] is a wildcard-free match, so a priority added to the shared
/// vocabulary fails to compile until it is placed there, and this crate's suite reconciles
/// this list and [`PriorityMappingConfig`]'s members against that enum's own derived schema.
pub const PRIORITY_LEVELS: [Priority; 4] = [
    Priority::Urgent,
    Priority::High,
    Priority::Medium,
    Priority::Low,
];

/// Where one priority sits in [`PRIORITY_LEVELS`], or `None` for `none`, which is no option;
/// see that list for what this pins.
#[must_use]
pub const fn level_position(priority: Priority) -> Option<usize> {
    match priority {
        Priority::None => None,
        Priority::Urgent => Some(0),
        Priority::High => Some(1),
        Priority::Medium => Some(2),
        Priority::Low => Some(3),
    }
}

/// This instance's complete priority-to-option mapping, read in both directions.
///
/// One option per level, held in [`PRIORITY_LEVELS`] order, once it is established that no
/// two levels name one option.
#[derive(Debug, Clone)]
struct PriorityMapping {
    options: [PriorityOptionName; 4],
}

impl PriorityMapping {
    fn resolve(config: PriorityMappingConfig, instance: &SourceName) -> Result<Self, SourceError> {
        let shipped = |name: &str| PriorityOptionName(name.to_owned());
        let mapping = Self {
            options: [
                config.urgent.unwrap_or_else(|| shipped("Urgent")),
                config.high.unwrap_or_else(|| shipped("High")),
                config.medium.unwrap_or_else(|| shipped("Medium")),
                config.low.unwrap_or_else(|| shipped("Low")),
            ],
        };
        for (index, option) in mapping.options.iter().enumerate() {
            if let Some(earlier) = mapping.options[..index]
                .iter()
                .position(|other| other.as_str().eq_ignore_ascii_case(option.as_str()))
            {
                return Err(SourceError::Config {
                    message: format!(
                        "priority_mapping of source {instance} sends both {} and {} to the board \
                         option {:?}; one option cannot read back as two priorities",
                        PRIORITY_LEVELS[earlier],
                        PRIORITY_LEVELS[index],
                        option.as_str()
                    ),
                });
            }
        }
        Ok(mapping)
    }

    /// The option `priority` lands on, or `None` for `none`, which is no option at all.
    fn option(&self, priority: Priority) -> Option<&str> {
        level_position(priority).map(|index| self.options[index].as_str())
    }

    /// The priority a board option name reports, or `None` when nothing maps to it.
    fn priority_of(&self, option: &str) -> Option<Priority> {
        self.options
            .iter()
            .position(|name| name.as_str().eq_ignore_ascii_case(option))
            .map(|index| PRIORITY_LEVELS[index])
    }

    /// Every mapped option name, in the order a new `Priority` field lists them.
    fn names(&self) -> impl Iterator<Item = &str> {
        self.options.iter().map(PriorityOptionName::as_str)
    }
}

/// One metadata value this source projects onto a board text field.
#[derive(Debug, Clone, Default, PartialEq, Eq, Deserialize, Serialize, schemars::JsonSchema)]
#[serde(deny_unknown_fields)]
pub struct MetadataFieldConfig {
    /// The name of the board text field the value is written to.
    pub field: String, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `MetadataField::resolve` refuses a blank, reserved, GitHub-owned or repeated name before the private validated entry is built.
    /// The top-level metadata key the value is read from. A key under the reserved
    /// `onetaskgraph.` prefix is refused.
    pub key: String, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `MetadataField::resolve` refuses a blank or reserved key before the private validated entry is built.
    /// Object keys walked inside that key's value, outermost first. Absent or empty, the
    /// key's own value is the one projected. Left out of a serialized entry when empty.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    // Kept in the schema as `"default": []`, for the reason `metadata_fields`' is.
    #[schemars(!skip_serializing_if)]
    pub path: Vec<String>,
}

/// The board fields no metadata value may be projected onto, compared ignoring case: the
/// three this source writes itself, and those GitHub owns on every board.
// llmlint: ignore[contracts_have_one_source_or_a_drift_gate] This list is the configuration contract `metadata_fields`' consumers agreed, refused at read so a mistaken entry is named early — not a mirror GitHub's field set has to stay in step with. What keeps a write off a field GitHub owns is the write's own check, `projection_field`, which refuses any field whose `dataType` is not `TEXT`: every built-in board field reports its own type (TITLE, ASSIGNEES, LABELS, …), and the live contract introspection holds GitHub to `ProjectV2Field.dataType`. So a built-in field GitHub adds later and this list lacks is still never written.
const UNPROJECTABLE_FIELDS: [&str; 13] = [
    STATUS_FIELD,
    PRIORITY_FIELD,
    ORIGIN_FIELD,
    "Title",
    "Assignees",
    "Labels",
    "Linked pull requests",
    "Milestone",
    "Repository",
    "Reviewers",
    "Parent issue",
    "Sub-issues progress",
    "Type",
];

/// One validated [`MetadataFieldConfig`] entry.
#[derive(Debug, Clone)]
struct MetadataField {
    field: ProjectedField,
    key: ProjectedKey,
    path: Vec<String>,
}

/// The name of a board text field a metadata value is projected onto: never blank, never one
/// this source or GitHub writes, and never one another entry of the same source names.
#[derive(Debug, Clone, PartialEq, Eq)]
struct ProjectedField(String);

/// The top-level metadata key a projected value is read from: never blank, and never under
/// the namespace this product reserves.
#[derive(Debug, Clone, PartialEq, Eq)]
struct ProjectedKey(String);

impl std::ops::Deref for ProjectedField {
    type Target = str;
    fn deref(&self) -> &str {
        &self.0
    }
}

impl std::ops::Deref for ProjectedKey {
    type Target = str;
    fn deref(&self) -> &str {
        &self.0
    }
}

/// What one item's metadata says one projected field should hold.
#[derive(Debug, Clone, PartialEq, Eq)]
enum Projected {
    /// A string to hold.
    Text(String),
    /// Nothing: the key or a step of the path is absent, or the value is `null`.
    Nothing,
}

impl MetadataField {
    /// Validate every entry of one instance's `metadata_fields`, refusing the first that
    /// cannot stand, naming the source and the entry.
    fn resolve(
        configured: Vec<MetadataFieldConfig>,
        instance: &SourceName,
    ) -> Result<Vec<Self>, SourceError> {
        let mut resolved: Vec<Self> = Vec::with_capacity(configured.len());
        for (index, entry) in configured.into_iter().enumerate() {
            let refuse = |why: String| SourceError::Config {
                message: format!(
                    "metadata_fields[{index}] of source {instance} (field {:?}, key {:?}) {why}",
                    entry.field, entry.key
                ),
            };
            if entry.field.trim().is_empty() {
                return Err(refuse("names a blank field".to_owned()));
            }
            if entry.key.trim().is_empty() {
                return Err(refuse("names a blank key".to_owned()));
            }
            if entry.key.split('.').next() == Some(MetadataKey::RESERVED_NAMESPACE) {
                return Err(refuse(format!(
                    "names a key under the reserved \"{}.\" prefix, which this product keeps \
                     for its own keys; next: project a caller key instead",
                    MetadataKey::RESERVED_NAMESPACE
                )));
            }
            if let Some(owned) = UNPROJECTABLE_FIELDS
                .iter()
                .find(|owned| owned.eq_ignore_ascii_case(&entry.field))
            {
                return Err(refuse(format!(
                    "names the board field {owned:?}, which this source or GitHub already \
                     writes; next: name a text field of your own"
                )));
            }
            if let Some(earlier) = resolved
                .iter()
                .position(|other| other.field.eq_ignore_ascii_case(&entry.field))
            {
                return Err(refuse(format!(
                    "names the same board field as metadata_fields[{earlier}]; one field \
                     cannot hold two values"
                )));
            }
            resolved.push(Self {
                field: ProjectedField(entry.field),
                key: ProjectedKey(entry.key),
                path: entry.path,
            });
        }
        Ok(resolved)
    }

    /// The path as a refusal spells it: `[]` for the key's own value.
    fn spelled_path(&self) -> String {
        serde_json::to_string(&self.path).unwrap_or_else(|_| format!("{:?}", self.path))
    }

    /// What `metadata` says this field should hold, or the refusal for a value no text field
    /// can hold.
    fn projected(
        &self,
        metadata: &BTreeMap<String, Value>,
        instance: &SourceName,
    ) -> Result<Projected, SourceError> {
        let mut value = metadata.get(&*self.key);
        for step in &self.path {
            value = value.and_then(|held| held.get(step.as_str()));
        }
        let found = match value {
            None | Some(Value::Null) => return Ok(Projected::Nothing),
            Some(Value::String(text)) => return Ok(Projected::Text(text.clone())),
            Some(Value::Bool(_)) => "a boolean",
            Some(Value::Number(_)) => "a number",
            Some(Value::Array(_)) => "an array",
            Some(Value::Object(_)) => "an object",
        };
        Err(SourceError::Refused {
            message: format!(
                "source {instance} projects metadata key {:?} at path {} onto its board text \
                 field {:?}, and this item holds {found} there; only a string, null or nothing \
                 can be written to a text field; next: store a string there, or remove the \
                 value",
                &*self.key,
                self.spelled_path(),
                &*self.field
            ),
        })
    }
}

/// What one item's `Priority` field says, read through this instance's mapping.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, schemars::JsonSchema)]
enum HeldPriority {
    /// A priority this source reports: an option the mapping names, or no value (`none`).
    Read(Priority),
    /// An option the mapping does not name, which is never read as a level or as `none`.
    Unmapped(String),
}

/// How fast this source writes, and how long it waits out a rate-limit refusal.
///
/// Configurable because a GitHub Enterprise installation sets its own limits and an
/// operator who has already been refused may want to go slower still — not because the
/// defaults are guesses.
#[derive(Debug, Clone, Default, PartialEq, Eq, Deserialize, Serialize, schemars::JsonSchema)]
#[serde(default, deny_unknown_fields)]
pub struct PacingConfig {
    /// Shortest interval between two content-creating mutations, in milliseconds.
    ///
    /// Zero sends them as fast as they are asked for, which is what a fixture server on
    /// loopback wants and what no board on github.com does. At most [`MAX_PACING_MS`].
    pub min_mutation_interval_ms: Option<u64>, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `Pacing::resolve` bounds it to `MAX_PACING_MS` before the private validated `Pacing` is built.
    /// First wait when a rate-limit refusal carries no hint, in milliseconds. Each
    /// further wait of the same call doubles it. At most [`MAX_PACING_MS`], and never
    /// zero while there is a budget to spend, because a schedule of zero-length waits
    /// consumes none of it and so never ends.
    pub retry_backoff_ms: Option<u64>, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `Pacing::resolve` refuses a non-progressing zero and bounds the rest before the private validated `Pacing` is built.
    /// Total time one call may spend waiting out rate limits, in milliseconds.
    ///
    /// Zero reports the refusal rather than waiting at all. At most [`MAX_PACING_MS`]:
    /// the bound is what makes this a wait rather than a hang.
    pub retry_budget_ms: Option<u64>, // llmlint: ignore[invalid_states_unrepresentable] Schema DTO; `Pacing::resolve` bounds it to `MAX_PACING_MS` before the private validated `Pacing` is built.
}

/// The largest any pacing setting may be, in milliseconds.
///
/// One hour. GitHub's own harshest published bound on content-generating requests works
/// out at one every 7.2 seconds, so an hour is already three orders of magnitude past
/// anything a real limit asks for, and past it the settings stop describing pacing at all:
/// a wait budget beyond it is the unbounded wait this whole mechanism exists to replace,
/// and an interval beyond it is a command that never sends its second mutation. It also
/// keeps the clock arithmetic in [`GitHubProjectsSource::reserve_mutation_slot`] inside
/// what a `Duration` can hold on every platform.
pub const MAX_PACING_MS: u64 = 3_600_000;

/// [`PacingConfig`] with every default resolved and every value checked, which is what the
/// source holds.
///
/// Unset, `min_mutation_interval_ms` spaces nothing, and every request is paced by GitHub's
/// secondary windows alone — see the `pacing` module; set, it spaces content-creating mutations
/// on top of those windows, [`MIN_MUTATION_INTERVAL_MS`] being the spacing that holds a sustained
/// burst under the per-minute bound by spacing alone; and zero sends every request as fast as it
/// is asked for, windows included.
#[derive(Debug, Clone, Copy)]
struct Pacing {
    /// Whether requests are held for GitHub's secondary windows: always, unless
    /// `min_mutation_interval_ms` is set to exactly zero, which has always asked for no
    /// pacing at all — what a fixture server on loopback wants and no board on github.com does.
    windows: bool,
    min_mutation_interval: Duration,
    retry_backoff: Duration,
    retry_budget: Duration,
}

impl Pacing {
    /// Resolve one instance's pacing, refusing a configuration that would not pace at all.
    fn resolve(config: PacingConfig, instance: &SourceName) -> Result<Self, SourceError> {
        let bounded = |value: Option<u64>, default: u64, field: &str| match value {
            Some(value) if value > MAX_PACING_MS => Err(SourceError::Config {
                message: format!(
                    "pacing.{field} of source {instance} is {value} ms, and the most any pacing \
                     setting may be is {MAX_PACING_MS} ms — an hour, which is already far past \
                     GitHub's own harshest published limit"
                ),
            }),
            Some(value) => Ok(Duration::from_millis(value)),
            None => Ok(Duration::from_millis(default)),
        };
        let retry_backoff = bounded(
            config.retry_backoff_ms,
            RETRY_BACKOFF_MS,
            "retry_backoff_ms",
        )?;
        let retry_budget = bounded(config.retry_budget_ms, RETRY_BUDGET_MS, "retry_budget_ms")?;
        if retry_backoff.is_zero() && !retry_budget.is_zero() {
            return Err(SourceError::Config {
                message: format!(
                    "pacing.retry_backoff_ms of source {instance} is 0 while \
                     pacing.retry_budget_ms is {} ms; a schedule of zero-length waits spends \
                     none of that budget, so it would retry a refusal forever. Set a backoff of \
                     at least 1 ms, or set retry_budget_ms to 0 to report a refusal without \
                     waiting at all",
                    retry_budget.as_millis()
                ),
            });
        }
        Ok(Self {
            windows: config.min_mutation_interval_ms != Some(0),
            min_mutation_interval: bounded(
                config.min_mutation_interval_ms,
                0,
                "min_mutation_interval_ms",
            )?,
            retry_backoff,
            retry_budget,
        })
    }
}

/// Factory for [`GitHubProjectsSource`].
#[derive(Debug, Clone, Copy, Default)]
pub struct Plugin;

impl SourcePlugin for Plugin {
    fn kind(&self) -> &'static str {
        KIND
    }
    /// The endpoint, the owner and the board's number: what names one board.
    fn remote_identity(&self, config: &Value) -> Option<String> {
        let config: GitHubProjectsConfig = serde_json::from_value(config.clone()).ok()?;
        Some(format!(
            "{}\n{}\n{}",
            config.endpoint, config.owner, config.project_number
        ))
    }
    fn config_schema(&self) -> Schema {
        schema_for!(GitHubProjectsConfig)
    }
    fn build(
        &self,
        name: &SourceName,
        config: &Value,
        secrets: &dyn SecretResolver,
    ) -> Result<Box<dyn TaskSource>, SourceError> {
        self.build_recording_into(name, config, secrets, Arc::new(Accounting::new()))
    }

    fn build_with_clock(
        &self,
        name: &SourceName,
        config: &Value,
        secrets: &dyn SecretResolver,
        clock: SharedClock,
    ) -> Result<Box<dyn TaskSource>, SourceError> {
        self.build_recording_with_clock(name, config, secrets, Arc::new(Accounting::new()), clock)
    }
}

/// The client `cell` holds, building it with `build` the first time it is asked for.
///
/// A failed build is the request's failure, with the message the source's construction gave it
/// before clients were deferred.
// llmlint: ignore[changed_behavior_has_e2e] A client build fails only when the TLS backend cannot initialise, which nothing outside the process can induce: a certificate bundle of garbage still builds one. That the build is deferred is proven end to end by `store_startup::a_command_builds_no_client_for_a_source_it_sends_nothing`.
fn built(
    cell: &std::sync::OnceLock<Client>,
    build: impl FnOnce() -> Result<Client, SourceError>,
) -> Result<&Client, SourceError> {
    if let Some(client) = cell.get() {
        return Ok(client);
    }
    let client = build()?;
    Ok(cell.get_or_init(|| client))
}

impl Plugin {
    /// Build a source recording every request it sends into an accounting the caller holds.
    ///
    /// [`SourcePlugin::build`] is this with an accounting of its own, which is what the
    /// registry gets. This is for a caller that is also calling GitHub itself and wants one
    /// session total rather than two — see [`accounting`] and
    /// [`GitHubProjectsSource::recording_into`].
    ///
    /// # Errors
    ///
    /// Exactly [`SourcePlugin::build`]'s, with the same source name in front of each:
    /// [`SourceError::Config`] for configuration this plugin cannot use and
    /// [`SourceError::Auth`] for a credential it cannot find.
    pub fn build_recording_into(
        &self,
        name: &SourceName,
        config: &Value,
        secrets: &dyn SecretResolver,
        ledger: Arc<Accounting>,
    ) -> Result<Box<dyn TaskSource>, SourceError> {
        self.build_recording_with_clock(name, config, secrets, ledger, system_clock())
    }

    fn build_recording_with_clock(
        &self,
        name: &SourceName,
        config: &Value,
        secrets: &dyn SecretResolver,
        ledger: Arc<Accounting>,
        clock: SharedClock,
    ) -> Result<Box<dyn TaskSource>, SourceError> {
        let config: GitHubProjectsConfig =
            serde_json::from_value(config.clone()).map_err(|e| SourceError::Config {
                message: format!("source {name}: {e}"),
            })?;
        let prefix = format!("source {name}: ");
        let mut source = GitHubProjectsSource::recording_into(name, config, secrets, ledger)
            .map_err(|error| match error {
                // The shared `StatusMapping::distinct` names the source itself.
                SourceError::Config { message } if message.starts_with(&prefix) => {
                    SourceError::Config { message }
                }
                SourceError::Config { message } => SourceError::Config {
                    message: format!("{prefix}{message}"),
                },
                SourceError::Auth { message } => SourceError::Auth {
                    message: format!("source {name}: {message}"),
                },
                other => other,
            })?;
        source.clock = clock;
        Ok(Box::new(source))
    }
}

/// Where a status category lands on this board, once configuration is resolved.
#[derive(Debug, Clone, PartialEq, Eq)]
enum StatusTarget {
    /// Not usable against this instance for this kind, and why.
    Disabled(UnmappedStatus),
    /// The board's `Status` option of this name.
    Column(ColumnName),
    /// A closed issue, with both its board option and the reason that says which closed it means.
    // llmlint: ignore[invalid_states_unrepresentable] The reason is fixed by the category — `done` closes as completed, `cancelled` as not planned — and this private enum is built in one place, `BoardStatuses::resolve`, which pairs each from the category's own slot. Carrying the reason on the target is what lets every write site that holds only a target derive its `stateInput` from that one resolved model rather than re-deriving it from a category and risking a disagreement with the mapping.
    Terminal(ColumnName, ClosedState),
}

/// Every status category, in the order the vocabulary declares them.
///
/// This list mirrors `StatusCategory`, so it carries its own drift gate rather than a
/// reviewer's attention: [`category_position`] is a wildcard-free match, so a variant
/// added to the shared vocabulary fails to compile until it is named there, and this
/// crate's suite reconciles this list against that enum's own derived schema, which is
/// generated from the variants rather than written beside them. The schema is what
/// catches a list left one short — a list checking only the positions it already holds
/// would pass while every mapping indexed by the new position panicked.
pub const CATEGORIES: [StatusCategory; 8] = [
    StatusCategory::Draft,
    StatusCategory::Backlog,
    StatusCategory::Todo,
    StatusCategory::Queued,
    StatusCategory::InProgress,
    StatusCategory::Done,
    StatusCategory::Cancelled,
    StatusCategory::Unknown,
];

/// Where one category sits in [`CATEGORIES`]; see that list for what this pins.
#[must_use]
pub const fn category_position(category: StatusCategory) -> usize {
    match category {
        StatusCategory::Draft => 0,
        StatusCategory::Backlog => 1,
        StatusCategory::Todo => 2,
        StatusCategory::Queued => 3,
        StatusCategory::InProgress => 4,
        StatusCategory::Done => 5,
        StatusCategory::Cancelled => 6,
        StatusCategory::Unknown => 7,
    }
}

/// The spelling a status category is configured and reported under.
fn category_name(category: StatusCategory) -> &'static str {
    match category {
        StatusCategory::Draft => "draft",
        StatusCategory::Backlog => "backlog",
        StatusCategory::Todo => "todo",
        StatusCategory::Queued => "queued",
        StatusCategory::InProgress => "in-progress",
        StatusCategory::Done => "done",
        StatusCategory::Cancelled => "cancelled",
        StatusCategory::Unknown => "unknown",
    }
}

/// A shipped default's option name.
///
/// The literals below are this file's own and non-blank, and they are validated by the
/// one constructor a configured name goes through rather than beside it.
fn shipped_column(name: &'static str) -> ColumnName {
    ColumnName::try_from(name.to_owned()).expect("a shipped default names a board option")
}

/// The shipped default for one category this instance's `status_mapping` does not mention,
/// for either kind.
fn shipped_default(category: StatusCategory) -> StatusTarget {
    match category {
        StatusCategory::Backlog => StatusTarget::Column(shipped_column("Backlog")),
        StatusCategory::Todo => StatusTarget::Column(shipped_column("Todo")),
        StatusCategory::Queued => StatusTarget::Column(shipped_column("Queued")),
        StatusCategory::InProgress => StatusTarget::Column(shipped_column("In Progress")),
        StatusCategory::Done => {
            StatusTarget::Terminal(shipped_column("Done"), ClosedState::Completed)
        }
        StatusCategory::Cancelled => {
            StatusTarget::Terminal(shipped_column("Cancelled"), ClosedState::NotPlanned)
        }
        StatusCategory::Draft | StatusCategory::Unknown => {
            StatusTarget::Disabled(UnmappedStatus::Unconfigured)
        }
    }
}

/// The two kinds a status is written and read for, each with its own half of the mapping.
const STATUS_KINDS: [ItemKind; 2] = [ItemKind::Task, ItemKind::Project];

/// This instance's complete category-to-target mapping for each kind, read in both
/// directions.
///
/// One target per category per kind, held at that category's own [`category_position`], so
/// a category missing from the mapping, named twice in it, or filed out of order is a state
/// this type cannot hold rather than one [`Self::target`] has to defend against. Both kinds'
/// targets are options of the board's one `Status` field.
#[derive(Debug, Clone)]
struct BoardStatuses {
    tasks: [StatusTarget; CATEGORIES.len()],
    projects: [StatusTarget; CATEGORIES.len()],
}

impl BoardStatuses {
    /// Resolve `configured` against the shipped defaults, refusing two categories one kind
    /// would read back from one option.
    ///
    /// A category the mapping does not mention keeps its shipped default for both kinds; one
    /// it does mention is exactly what it configures, so a per-kind object leaves the kind it
    /// omits unmapped rather than defaulted.
    fn resolve(configured: &StatusMapping, instance: &SourceName) -> Result<Self, SourceError> {
        let resolve_kind =
            |kind: ItemKind| -> Result<[StatusTarget; CATEGORIES.len()], SourceError> {
                // `CATEGORIES[position] == category` for every category — the crate's suite
                // asserts it — so mapping the list in order fills each category's own slot.
                let mut targets = CATEGORIES.map(shipped_default);
                for (slot, category) in targets.iter_mut().zip(CATEGORIES) {
                    if !configured.mentions(category) {
                        continue;
                    }
                    *slot = match configured.name_for(category, kind) {
                        Err(why) => StatusTarget::Disabled(why),
                        Ok(name) => {
                            let option = ColumnName::try_from(name.as_str().to_owned())
                                .map_err(|message| SourceError::Config { message })?;
                            match category {
                                StatusCategory::Done => {
                                    StatusTarget::Terminal(option, ClosedState::Completed)
                                }
                                StatusCategory::Cancelled => {
                                    StatusTarget::Terminal(option, ClosedState::NotPlanned)
                                }
                                _ => StatusTarget::Column(option),
                            }
                        }
                    };
                }
                StatusMapping::distinct(
                    instance,
                    kind,
                    CATEGORIES
                        .iter()
                        .zip(&targets)
                        .filter_map(|(category, target)| target.option().map(|o| (*category, o))),
                )?;
                Ok(targets)
            };
        Ok(Self {
            tasks: resolve_kind(ItemKind::Task)?,
            projects: resolve_kind(ItemKind::Project)?,
        })
    }

    /// Every category's target for `kind`, in category order.
    const fn targets(&self, kind: ItemKind) -> &[StatusTarget; CATEGORIES.len()] {
        match kind {
            ItemKind::Task => &self.tasks,
            ItemKind::Project => &self.projects,
        }
    }

    fn target(&self, kind: ItemKind, category: StatusCategory) -> &StatusTarget {
        &self.targets(kind)[category_position(category)]
    }

    /// The category a board option name reports for `kind`, or `None` when nothing of that
    /// kind maps to it.
    fn category_of(&self, kind: ItemKind, option: &str) -> Option<StatusCategory> {
        CATEGORIES.into_iter().find(|category| {
            self.target(kind, *category)
                .option()
                .is_some_and(|name| name.eq_ignore_ascii_case(option))
        })
    }

    /// Every option name either kind maps a category to, each once ignoring case, in
    /// category order with a task's name before a project's — what the guarded setup asks
    /// the `Status` field to hold.
    fn wanted(&self) -> Vec<String> {
        let mut wanted: Vec<String> = Vec::new();
        for category in CATEGORIES {
            for kind in STATUS_KINDS {
                if let Some(name) = self.target(kind, category).option()
                    && !wanted.iter().any(|held| held.eq_ignore_ascii_case(name))
                {
                    wanted.push(name.to_owned());
                }
            }
        }
        wanted
    }

    /// The status an item of `kind` reports, from the three things a read of it says: its
    /// board `Status` option, whether its issue is closed, and the reason it was closed with.
    ///
    /// **A closed issue's category follows its closure**, by [`closed_category`]: completed
    /// or no reason reads `done`, not planned and duplicate read `cancelled`, and any other
    /// reason reads `unknown`. `REOPENED`-while-closed is a state this source can never
    /// produce, so it is read permissively rather than refused — reads are faithful, and
    /// refusals belong on writes.
    ///
    /// **Its name follows the closure too, unless the option agrees with it.** The option's
    /// own name is kept when this kind's mapping reads that option as the same category, or
    /// does not name it at all — so a closed issue in a "Shipped" column reports `done`
    /// named `Shipped`. An option the mapping reads as another category contradicts the
    /// closure, and so does no option at all: the name is then the option the mapping gives
    /// the closure's category, so an issue closed as completed at `Todo` reads `done` named
    /// by the board's `done` option, and a destination reading `Todo` as todo is never handed
    /// it. An `unknown` closure keeps the option, or `Closed` without one.
    ///
    /// An open item's option reads through its own kind's mapping, and an option that mapping
    /// does not name reads as `Unknown` under its own name.
    ///
    /// One function of those three rather than of a response, so a narrow status write can
    /// answer what a re-read would report by applying it to the state it has just written.
    fn status(
        &self,
        kind: ItemKind,
        option: Option<&str>,
        closed: bool,
        reason: Option<&str>,
    ) -> Status {
        if closed {
            let category = closed_category(reason);
            let agrees = option.filter(|option| {
                category == StatusCategory::Unknown
                    || self
                        .category_of(kind, option)
                        .is_none_or(|read| read == category)
            });
            let name = agrees.or_else(|| self.target(kind, category).option());
            let fallback = match category {
                StatusCategory::Done => "Done",
                StatusCategory::Cancelled => "Cancelled",
                _ => "Closed",
            };
            return Status {
                category,
                name: name.or(option).unwrap_or(fallback).to_owned(),
            };
        }
        let name = option.unwrap_or("Open").to_owned();
        Status {
            category: self
                .category_of(kind, &name)
                .unwrap_or(StatusCategory::Unknown),
            name,
        }
    }

    /// The option a closed item's `Status` has to hold for its card to agree with its
    /// closure, and the category that closure reads as — or `None` when the item is open,
    /// its closure reads `unknown`, or its option already agrees.
    ///
    /// What `task status reconcile` writes: only `done` and `cancelled` closures are brought
    /// in line, each to the one option the mapping gives its category — an item with no
    /// option, or at one the mapping does not name, included.
    fn contradicted(
        &self,
        kind: ItemKind,
        option: Option<&str>,
        closed: bool,
        reason: Option<&str>,
    ) -> Option<(ClosedAs, &str)> {
        if !closed {
            return None;
        }
        let closed_as = match closed_category(reason) {
            StatusCategory::Done => ClosedAs::Done,
            StatusCategory::Cancelled => ClosedAs::Cancelled,
            _ => return None,
        };
        let wanted = self.target(kind, closed_as.category()).option()?;
        if option.is_some_and(|option| option.eq_ignore_ascii_case(wanted)) {
            return None;
        }
        Some((closed_as, wanted))
    }
}

/// The category a closed issue reads as, from the reason it was closed with.
///
/// `COMPLETED`, and no reason at all, is finished work; `NOT_PLANNED` and `DUPLICATE` are both
/// work nobody will do here — a duplicate's work is the other issue's — and any other reason
/// is one this source cannot read a category from.
fn closed_category(reason: Option<&str>) -> StatusCategory {
    match reason {
        None | Some("COMPLETED") => StatusCategory::Done,
        Some("NOT_PLANNED" | "DUPLICATE") => StatusCategory::Cancelled,
        Some(_) => StatusCategory::Unknown,
    }
}

impl BoardStatuses {
    /// For each kind, the option names it maps a category to that `existing` lacks, ignoring
    /// case; a kind lacking none is left out.
    fn missing_by_kind(&self, existing: &[StatusOption]) -> Vec<KindMissing> {
        STATUS_KINDS
            .into_iter()
            .filter_map(|kind| {
                let missing: Vec<String> = self
                    .targets(kind)
                    .iter()
                    .filter_map(StatusTarget::option)
                    .filter(|wanted| {
                        !existing
                            .iter()
                            .any(|option| option.name.as_str().eq_ignore_ascii_case(wanted))
                    })
                    .map(str::to_owned)
                    .collect();
                (!missing.is_empty()).then_some(KindMissing { kind, missing })
            })
            .collect()
    }
}

impl StatusTarget {
    /// The board option this target selects, or `None` for an unmapped one.
    fn option(&self) -> Option<&str> {
        match self {
            Self::Column(name) | Self::Terminal(name, _) => Some(name.as_str()),
            Self::Disabled(_) => None,
        }
    }
}

// llmlint: ignore-block[comments_earn_their_place, contracts_have_one_source_or_a_drift_gate] Every `createIssue` names one of these, and which one is the rule — a reader who reaches the type from `create_and_file_issue` gets the rule in one sentence here without the method's refusals, which stay on `creation_target`, the rule's one executable source; `tests/plugin.rs` drives every arm of it against the loopback board.
/// One repository this source can create an issue in, as `owner/name`.
///
/// Every `createIssue` this source sends names one of these: the item's own single
/// `repositories` entry — on a source declared private, only when it reads private — else
/// its parent project issue's repository, else the configured
/// [`GitHubProjectsConfig::repository`]. [`GitHubProjectsSource::creation_target`] makes
/// that choice and says what it refuses before `createIssue`.
// llmlint: ignore-end[comments_earn_their_place, contracts_have_one_source_or_a_drift_gate]
#[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize)]
#[serde(try_from = "String", into = "String")]
struct RepositoryTarget {
    owner: String, // llmlint: ignore[invalid_states_unrepresentable] Private, constructed only after `owner/name` validation in `new`.
    name: String, // llmlint: ignore[invalid_states_unrepresentable] Private, constructed only after `owner/name` validation in `new`.
}

impl RepositoryTarget {
    fn parse(value: &str) -> Result<Self, SourceError> {
        let (owner, name) = value.split_once('/').ok_or_else(|| SourceError::Config {
            message: format!(
                "repository must be spelled owner/name; {value:?} names no repository"
            ),
        })?;
        if !valid_github_owner(owner) || !valid_github_repository_name(name) {
            return Err(SourceError::Config {
                message: format!(
                    "repository must be spelled owner/name with a GitHub login and one \
                     repository name; {value:?} is not"
                ),
            });
        }
        Ok(Self {
            owner: owner.to_owned(),
            name: name.to_owned(),
        })
    }

    /// The one host whose repositories this source creates issues in, spelled once: it is
    /// what [`Self::origin`] renders and what [`Self::from_origin`] accepts.
    const HOST: &str = "github.com";

    fn origin(&self) -> String {
        format!("{}/{}/{}", Self::HOST, self.owner, self.name)
    }

    /// The repository a normalized origin names, or why it is none this source can create
    /// an issue in: another host, or more or fewer than `owner/name` under this one.
    fn from_origin(origin: &Repository) -> Result<Self, String> {
        let not_here = || {
            format!(
                "{} is not a {}/owner/name repository",
                origin.as_str(),
                Self::HOST
            )
        };
        let (host, rest) = origin.as_str().split_once('/').ok_or_else(not_here)?;
        if host != Self::HOST {
            return Err(not_here());
        }
        Self::parse(rest).map_err(|_| not_here())
    }

    fn slug(&self) -> String {
        format!("{}/{}", self.owner, self.name)
    }
}

/// Read back from a kept context as the `owner/name` it was written as, refused unless
/// [`RepositoryTarget::parse`] takes it.
impl TryFrom<String> for RepositoryTarget {
    type Error = String;

    fn try_from(slug: String) -> Result<Self, Self::Error> {
        Self::parse(&slug).map_err(|error| error.to_string())
    }
}

impl From<RepositoryTarget> for String {
    fn from(target: RepositoryTarget) -> Self {
        target.slug()
    }
}

/// How long GitHub's issue search may go on missing a comment after it was made or edited,
/// which a read of the comments made since an instant sets its watermark back by: four times the
/// thirty seconds the credentialed lane has watched it lag.
pub const COMMENT_SEARCH_LAG: std::time::Duration = std::time::Duration::from_secs(120);

/// A source which reads GitHub afresh for every operation.
pub struct GitHubProjectsSource {
    /// This source's configured name, used both to tell a far end naming this source
    /// from one naming a system it knows nothing about, and to name the instance a
    /// status refusal is about.
    name: SourceName,
    owner: String, // llmlint: ignore[invalid_states_unrepresentable] Private, constructed only by `new` after full GitHub-owner validation.
    project_number: u32, // llmlint: ignore[invalid_states_unrepresentable] Private, constructed only by `new` after GraphQL-Int validation.
    repository: Option<RepositoryTarget>,
    /// The logins whose repositories a create may file in; empty restricts nothing.
    creation_owners: Vec<String>, // llmlint: ignore[invalid_states_unrepresentable] Private, constructed only by `new` after each entry passes GitHub's login grammar.
    endpoint: Url,
    token: SecretString,
    credential_name: String, // llmlint: ignore[invalid_states_unrepresentable] Private diagnostic value constructed only after environment-name validation.
    statuses: BoardStatuses,
    /// Where each priority lands on this board, or `None` when this instance holds none.
    priorities: Option<PriorityMapping>,
    /// The metadata values this instance projects onto board text fields, in configured order.
    metadata_fields: Vec<MetadataField>,
    /// What this board's sub-issues mean.
    sub_issues: SubIssues,
    /// The link types this source mirrors as `relates-to`.
    link_types: onetaskgraph_plugin_api::LinkTypes,
    /// Whether this source's configuration opted in to assignees: every document it sends is
    /// then spelled by [`graphql::with_assignees`], and every task it reads carries them.
    assignees: bool,
    /// The GitHub node id of each person an assignee write of this process has named, by
    /// lowercased login — `None` for a login GitHub said it does not know — so a later call
    /// naming one sends no lookup. A write that fails, or is answered with other people, has the
    /// people it named forgotten when its call ends ([`Self::forgotten`]), so the next call
    /// naming them asks GitHub again while the rest of its own call does not.
    people: Mutex<BTreeMap<String, Option<String>>>,
    /// The node ids of the people a failed or mismatched assignee write of this call named,
    /// forgotten from [`Self::people`] when the call ends.
    forgotten: Mutex<Vec<String>>,
    /// GitHub's own spelling of each login this process looked up, by lowercased login: what a
    /// write that assigned a person reports them as, however the caller spelled them.
    spellings: Mutex<BTreeMap<String, String>>,
    /// Built on the first request rather than with the source, as [`Self::client`] says.
    client: std::sync::OnceLock<Client>,
    /// Built on the first attachment request, as [`Self::asset_client`] says.
    asset_client: std::sync::OnceLock<Client>,
    /// The markup the engine handed this source for its bodies' headings, applied to every
    /// visible body it composes and stripped from every one it reads back: see
    /// [`TaskSource::attach_body_markup`].
    body_markup: std::sync::OnceLock<onetaskgraph_plugin_api::SharedBodyMarkup>,
    /// Every item this source has created in this command, in the order it created them —
    /// dropped by [`TaskSource::end_command`].
    ///
    /// GitHub's `projectV2.items` is eventually consistent: an issue added to a board with
    /// `addProjectV2ItemById` is routinely absent from the very next read of that board, so
    /// a copy resolving a dependency on an item it had just created refused it as not
    /// found. A board read is completed from this — an item remembered here and absent from
    /// the read is added back, because the board really does hold it and only the read is
    /// behind.
    ///
    /// It is not a cache of a user's work: nothing is remembered that this process did not
    /// itself just write, it lives and dies with the process, and it is never consulted for
    /// an item this source did not create.
    created: Mutex<Vec<Resolved>>,
    /// Every item that already existed and that this source has written in this command, as
    /// it wrote it — dropped by [`TaskSource::end_command`].
    ///
    /// The other half of [`Self::created`], held on the same terms and for the reason a
    /// narrowed read needs it: an answer from GitHub's search or from the board's own field
    /// filter is an index behind a write this process made moments ago, so a query matching
    /// what this process just wrote onto an existing item would otherwise miss it. Nothing
    /// is remembered that this process did not itself just write.
    updated: Mutex<Vec<Resolved>>,
    /// Every issue this source has added a comment to or edited a comment of in this command
    /// — dropped by [`TaskSource::end_command`].
    ///
    /// A comment-activity read is narrowed by GitHub's issue search, whose `updated:` index
    /// lags the write that moved an issue's `updatedAt`, and neither [`Self::created`] nor
    /// [`Self::updated`] is moved by a comment, so an issue this process had just commented
    /// on was missing from such a read — or ruled out by the `updatedAt` its own record held
    /// from before — until the index caught up. Each id here is a candidate of every such
    /// search-narrowed read, and wherever it is a candidate its comments are read rather than
    /// it being ruled out by a stale `updatedAt`; that read is of the issue's own node, so it
    /// is current. It holds ids alone: nothing of a comment is remembered. A comment another
    /// process wrote is still found only once the index has it.
    commented: Mutex<Vec<NativeId>>,
    /// How fast this source writes, and how long it waits out a refusal.
    pacing: Pacing,
    /// When the last content-creating mutation finished, or the moment the furthest-out
    /// reserved slot releases the next one, whichever is later — so the one after it can be
    /// spaced from that, when `pacing.min_mutation_interval_ms` asks for a spacing. See
    /// [`GitHubProjectsSource::paced`] for why completion rather than release is what it is
    /// measured from.
    last_mutation: Mutex<Option<Duration>>,
    /// The `Date` GitHub stated on the last answer this source read: its own time, which the
    /// watermark of a comment read is set back from. See [`TaskSource::served_at`].
    last_served: Mutex<Option<DateTime<Utc>>>,
    /// GitHub's secondary windows as this source has charged them; see the `pacing` module.
    windows: Mutex<pacing::Windows>,
    /// The tasks a project holds, as a copy's batched read of the project and every one of them
    /// found them — see [`TaskSource::read_ahead_for_copy`] — for the length of one command, so
    /// a listing of that project's tasks is answered without a read of its own.
    task_children: Mutex<BTreeMap<NativeId, Vec<Resolved>>>,
    /// The web address of each issue this source created in the current command, as its
    /// `createIssue` answered it — what [`TaskSource::written_location`] answers, without a
    /// request.
    written: Mutex<BTreeMap<NativeId, String>>,
    /// What a copy has held back of the items it created, between
    /// [`TaskSource::begin_deferred_writes`] and [`TaskSource::finish_writes`] — `None` while
    /// nothing is being held back, when every write lands whole as it is made.
    deferred: Mutex<Option<Deferred>>,
    /// Each existing issue's update held back while writes are, in the order it was made: sent
    /// together by [`Self::send_held_updates`] before any other request goes out, so nothing
    /// this source sends or reads ever sees the board without them.
    held_updates: Mutex<Vec<HeldUpdate>>,
    /// The first page of comments a comment-activity read read of each candidate it confirmed,
    /// held for the one comment listing of that task that follows in this command — see
    /// `batch::hold_comment_pages`.
    held_comments: Mutex<BTreeMap<NativeId, Page<Comment>>>,
    /// The board field writes and issue updates a batch of targeted updates is holding back,
    /// while one runs — see `batch::TaskWrites`.
    task_writes: Mutex<Option<batch::TaskWrites>>,
    /// The change token each task a targeted update wrote in the current command holds after
    /// that write, by the task's id with its board item's id — spelled from the write's own
    /// last response, or from the read made right after it where no response showed the whole
    /// issue: what [`TaskSource::written_task`] answers as its `version`. Any later mutation
    /// naming the task or its board item drops its entry before it is sent, so an entry is
    /// never one write behind.
    /// A response that showed the issue whole but not every part of its token — a draft, or a
    /// page of labels or field values that did not end — records `None`, which a read would
    /// not improve on.
    written_tokens: Mutex<BTreeMap<NativeId, (String, Option<String>)>>,
    clock: SharedClock,
    /// Whether the configuration declares this source private, as
    /// [`TaskSource::declare_visibility`] told it: what decides where a new issue whose one
    /// named repository reads public is created — see [`Self::creation_rule`].
    declared_private: std::sync::atomic::AtomicBool,
    numeric_repositories: tokio::sync::Mutex<BTreeMap<RepositoryTarget, std::num::NonZeroU64>>,
    /// The board as this process last read it, for the length of one command — dropped by
    /// [`TaskSource::end_command`].
    ///
    /// A copy of a project used to re-read the whole board, paged, before writing each of
    /// its items, which is by far the largest part of a copy's request count and none of
    /// its work. Nothing else changes this board while a command runs — this source's own
    /// writes are the only writer — so one read answers them all.
    ///
    /// It is not a store of a user's work and it is not the cache the no-persistence
    /// invariant forbids: it lives and dies with the process exactly as `created` does,
    /// nothing is written down, and [`Self::board`] still completes it from `created`, so
    /// an item this command created and then depends on resolves whether or not GitHub's
    /// own eventually-consistent read has caught up. A write to an item already on the
    /// board updates the entry here too, so what this holds is the last read plus this
    /// process's own writes rather than a snapshot taken before them.
    board_cache: Mutex<Option<Board>>,
    /// Every issue this board's own search reported, for the length of one command — dropped
    /// by [`TaskSource::end_command`].
    ///
    /// The second half of a board read, and cached for the same reason and on the same
    /// terms as the first: it lives and dies with the process, nothing is written down, and
    /// a write this process makes updates the entry here exactly as it updates the one in
    /// [`Self::board_cache`]. One read answers every question a command asks, so a command
    /// that lists this board's projects and its tasks pays for one search rather than two.
    search_cache: Mutex<Option<Vec<Resolved>>>,
    /// What each narrowed question GitHub was asked answered, keyed by that question, for
    /// the length of one command — dropped by [`TaskSource::end_command`].
    ///
    /// The narrowed counterpart of [`Self::search_cache`], held on the same terms: it lives
    /// and dies with the process, nothing is written down, a write this process makes
    /// updates the entry here as it updates the other two, and every answer is completed
    /// with this process's own writes each time it is given. A command that asks the same
    /// narrowed question twice — a wait polling for its own items, a listing repeated after a
    /// write — pays for it once, which is what the whole-board read it replaced gave it.
    narrowed_cache: Mutex<BTreeMap<String, Vec<Resolved>>>,
    search_next: Mutex<BTreeMap<String, Option<String>>>,
    /// Records already resolved in this command, reused by writes and for comment identity.
    /// Explicit item reads still reach GitHub. Nothing is persisted, and
    /// [`TaskSource::end_command`] drops every record, so a write in the next command reads
    /// its item as a person has since left it.
    resolved_cache: Mutex<BTreeMap<NativeId, Resolved>>,
    /// What [`TaskSource::read_ahead_for_copy`] read, by id — `None` for an id that read
    /// answered as holding nothing — each taken by the next read of that item by id and by
    /// nothing else, dropped by any mutation naming it, and cleared by
    /// [`TaskSource::end_command`]. So a read answered from it is the copy's own
    /// read-before-write, made in the same request as the others' rather than on its own.
    read_ahead: Mutex<BTreeMap<NativeId, Option<Resolved>>>,
    /// The origins [`TaskSource::read_ahead_origins_for_copy`] was asked about and has not sent
    /// yet: held for the [`TaskSource::read_ahead_for_copy`] that follows it, which sends them —
    /// beside the members of the one project it names, when it names one — or for the next
    /// request this source makes, whichever comes first.
    pending_origins: Mutex<Vec<String>>,
    /// Each project's members as a lookup carrying them read them in this command, trimmed to
    /// their classification and repositories, taken by [`TaskSource::member_classifications`]
    /// of that project and by nothing else, and cleared by [`TaskSource::end_command`].
    member_classes: Mutex<BTreeMap<NativeId, Vec<MemberClass>>>,
    /// Every item a listing of one project's members carried whole in this command — the
    /// project issue and each of its sub-issues — kept for [`TaskSource::read_ahead_for_copy`]
    /// alone, which takes each one it is asked for into [`Self::read_ahead`] instead of reading
    /// it again. No other read is answered from it, so a read by id outside a copy's
    /// read-before-write reads GitHub as it always did. Dropped by any mutation naming an
    /// item, and cleared by [`TaskSource::end_command`].
    carried: Mutex<BTreeMap<NativeId, Resolved>>,
    /// Items an issue read by reference reached, held for the one status write each was read
    /// for and taken out by it alone — see [`Self::take_held`] — because that read selects only
    /// what such a write needs. Dropped by [`TaskSource::end_command`].
    held_cache: Mutex<BTreeMap<NativeId, Resolved>>,
    /// Each project's sub-issues as GitHub answered them, keyed by the selector they were
    /// asked for under, with the project that selector named — for the length of one command,
    /// dropped by [`TaskSource::end_command`].
    ///
    /// A caller pages through a project's tasks one engine page at a time, and every page
    /// is cut from the whole list of them, so without this each page walked every
    /// [`graphql::SUB_ISSUES`] page again and a project of `n` listing pages cost `n²`
    /// requests to read once. Held on the terms [`Self::narrowed_cache`] is: it lives and
    /// dies with the process, nothing is written down, a write this process makes updates or
    /// removes the entry here as it does there, and every answer is completed with this
    /// process's own writes each time it is given.
    children_cache: Mutex<ProjectChildren>,
    /// The board's own id and field definitions as this process last read them on their
    /// own, for the length of one command — dropped by [`TaskSource::end_command`].
    ///
    /// What a write needs of the board and its item does not say, read once per command
    /// rather than once per item written, on the terms [`Self::board_cache`] is held on: it
    /// lives and dies with the process and nothing is written down. It holds no item and so
    /// can answer no question about one — see [`Self::board_fields`].
    fields_cache: Mutex<Option<BoardFields>>,
    /// Each destination repository's node id, resolved once per repository
    /// rather than per issue created.
    ///
    /// A repository's node id does not change, and re-reading it for every issue of a copy
    /// spent one request per item on an answer this source already had. It is a map rather
    /// than one entry because a copy files each item in the repository its own
    /// `repositories` field names, so a plan across five repositories asks GitHub five
    /// times and not once per item.
    repository_cache: Mutex<BTreeMap<RepositoryTarget, String>>,
    /// What every request this source sends is recorded into.
    ///
    /// Ordinary code path, not a mode: [`Self::send_once`] records into it at the one place
    /// a request leaves this crate, so nothing has to be switched on for a session to be
    /// counted. It is shared rather than owned so a caller accounting for a whole session —
    /// its own schema verification, board lookups, residue sweep and cleanup beside this
    /// source's reads and writes — adds up one accounting instead of two. See
    /// [`accounting`] for what a record carries and what a session's spend is and is not.
    ledger: Arc<Accounting>,
    /// What [`expected`] holds for a write the store's cache expects: the journal while one is
    /// open, the issues primed from the cache, and the comments known on each.
    expected: expected::Held,
    /// What [`projection`] holds for a producer's projection: a kept placement awaiting its
    /// setup-context token, and where the last create was placed.
    projection: projection::Held,
}

/// The board item a single field update's answer names, refused unless it is `item_id`.
fn field_update_item<'a>(data: &'a Value, item_id: &str) -> Result<&'a Value, SourceError> {
    let returned = data
        .pointer("/updateProjectV2ItemFieldValue/projectV2Item")
        .ok_or_else(|| SourceError::Malformed {
            message: "GitHub field update returned no project item".into(),
        })?;
    if required_str(returned, "id")? != item_id {
        return Err(SourceError::Malformed {
            message: "GitHub field update returned the wrong project item".into(),
        });
    }
    Ok(returned)
}

/// GitHub's closed single-select color vocabulary.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize, Serialize, schemars::JsonSchema)]
#[serde(rename_all = "SCREAMING_SNAKE_CASE")]
pub enum StatusOptionColor {
    /// Gray.
    Gray,
    /// Blue.
    Blue,
    /// Green.
    Green,
    /// Yellow.
    Yellow,
    /// Purple.
    Purple,
    /// Red.
    Red,
    /// Orange.
    Orange,
    /// Pink.
    Pink,
}

/// Whether a guarded board setup — of the fields, or of the Status options alone — plans or
/// applies its additions.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum SetupMode {
    /// Read without mutation.
    Plan,
    /// Apply and verify.
    Apply,
}

/// The name [`SetupMode`] had when Status was the one field set up, kept so a caller written
/// against it goes on compiling.
pub type StatusOptionsMode = SetupMode;

/// The explicit result of the requested operation.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, schemars::JsonSchema)]
#[serde(rename_all = "kebab-case")]
pub enum StatusOptionsOutcome {
    /// A read-only plan.
    Planned,
    /// Apply found nothing missing.
    Unchanged,
    /// Additions were applied and verified.
    Applied,
}

/// A GitHub single-select option's opaque GraphQL node identifier.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, schemars::JsonSchema)]
#[serde(transparent)]
pub struct StatusOptionId(#[schemars(length(min = 1))] String);

impl TryFrom<String> for StatusOptionId {
    type Error = String;

    fn try_from(id: String) -> Result<Self, Self::Error> {
        if id.trim().is_empty() {
            return Err("a GitHub Status option id cannot be blank".to_owned());
        }
        Ok(Self(id))
    }
}

/// One existing or proposed option in a guarded Status-field update.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, schemars::JsonSchema)]
pub struct StatusOption {
    /// GitHub's stable id.
    pub id: StatusOptionId,
    /// The visible option name.
    pub name: ColumnName,
    /// GitHub's single-select color token.
    pub color: StatusOptionColor,
    /// The option description, including an empty one.
    pub description: String,
}

/// One board item's Status assignment, retained as recovery data.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, schemars::JsonSchema)]
pub struct StatusAssignment {
    /// The project item id whose assignment this is.
    // llmlint: ignore[invalid_states_unrepresentable] This opaque GraphQL node ID is
    // carried verbatim as operator recovery data; introducing a semantic type would claim
    // validation rules GitHub does not publish and no operation here interprets.
    pub item_id: String,
    /// The selected option, absent when the item has no status.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub option: Option<AssignedStatusOption>,
}

/// The inseparable id and name of an assigned option.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, schemars::JsonSchema)]
pub struct AssignedStatusOption {
    /// GitHub's stable id.
    pub id: StatusOptionId,
    /// The visible name.
    pub name: ColumnName,
}

/// The plan and verified outcome of reconciling configured Status options.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, schemars::JsonSchema)]
pub struct StatusOptionsReport {
    /// The configured source name.
    pub source: SourceName,
    /// Configured option names absent before the operation.
    // llmlint: ignore[invalid_states_unrepresentable] Each value originates from a
    // `ColumnName` and has therefore already passed its nonblank validation; retaining the
    // serialized string here preserves the report's intentionally simple public contract.
    pub missing: Vec<String>,
    /// What the requested operation did.
    pub outcome: StatusOptionsOutcome,
    /// The complete option list observed before any mutation.
    pub existing: Vec<StatusOption>,
}

#[derive(Debug, Clone, PartialEq, Eq)]
struct StatusSnapshot {
    // llmlint: ignore[invalid_states_unrepresentable] This private opaque GraphQL ID is
    // passed back as the mutation's project identity; a newtype could enforce no stronger
    // invariant because GitHub publishes no grammar for it.
    board_id: String,
    // llmlint: ignore[invalid_states_unrepresentable] This private opaque GraphQL ID is
    // passed back as the mutation's field identity; a newtype could enforce no stronger
    // invariant because GitHub publishes no grammar for it.
    field_id: String,
    options: Vec<StatusOption>,
    assignments: Vec<StatusAssignment>,
}

/// The name of the board field a status is held in.
const STATUS_FIELD: &str = "Status";

/// Every item's value of each field `report` names, as it stood before the setup wrote
/// anything — what a person puts back when the setup is refused part way.
fn recovery(report: &FieldsReport, before: &BoardSnapshot) -> Result<String, SourceError> {
    let assignments: BTreeMap<&str, Vec<StatusAssignment>> = report
        .fields
        .iter()
        .map(|field| (field.field.name(), before.assignments(field.field)))
        .collect();
    serde_json::to_string_pretty(&assignments).map_err(|error| SourceError::Malformed {
        message: format!("cannot render the pre-write field recovery snapshot: {error}"),
    })
}

/// One board field the guarded setup reads and writes — every one it reads, and the only
/// ones it writes.
#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Serialize, schemars::JsonSchema)]
pub enum BoardField {
    /// The single-select `Status` field every instance's `status_mapping` resolves into.
    Status,
    /// The single-select `Priority` field an instance's `priority_mapping` resolves into.
    Priority,
}

impl BoardField {
    /// The field's name on the board.
    #[must_use]
    pub const fn name(self) -> &'static str {
        match self {
            Self::Status => STATUS_FIELD,
            Self::Priority => PRIORITY_FIELD,
        }
    }

    /// The field a board calls `name`, or `None` for one this setup does not own.
    fn named(name: &str) -> Option<Self> {
        [Self::Status, Self::Priority]
            .into_iter()
            .find(|field| field.name() == name)
    }
}

/// What the guarded setup did to one field.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, schemars::JsonSchema)]
#[serde(rename_all = "kebab-case")]
pub enum FieldOutcome {
    /// A read-only plan.
    Planned,
    /// Apply found the field there with every configured option.
    Unchanged,
    /// Missing options were added to the field that was there, and verified.
    Applied,
    /// The field was not there; it was created holding the configured options, and verified.
    Created,
}

/// One field's plan, or its verified outcome.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, schemars::JsonSchema)]
pub struct FieldReport {
    /// Which field.
    pub field: BoardField,
    /// Whether the board had the field before the operation.
    // llmlint: ignore[invalid_states_unrepresentable] `exists` beside `outcome` is the report's
    // wire shape as its consumer's contract fixes it — `{"field", "exists", "missing",
    // "outcome", "existing"}` — so folding one into the other would change a published JSON
    // shape. The contradictory pairings cannot be built: `GitHubProjectsSource::fields` is the
    // one constructor, and it derives `outcome` from `exists` in one match.
    pub exists: bool,
    /// Configured option names the field lacked before the operation — every one of them,
    /// in the order a new field lists them, when the field was not there at all.
    // llmlint: ignore[invalid_states_unrepresentable] Each value originates from a validated
    // mapping name and has therefore already passed its nonblank validation; the serialized
    // string is the report's intentionally simple public contract, as `StatusOptionsReport`'s is.
    pub missing: Vec<String>,
    /// For the `Status` field, which item kind each missing name is configured for: one
    /// entry per kind `status_mapping` names a missing option for, task before project, each
    /// listing that kind's missing names in category order. A name both kinds use is in
    /// both. Empty — and left out of the JSON — when nothing is missing, and always for
    /// `Priority`, which only a task holds.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    // Kept in the schema as `"default": []` although the JSON leaves an empty list out, so
    // both SDKs model an absent `kinds` as an empty list rather than as `null`.
    #[schemars(!skip_serializing_if)]
    pub kinds: Vec<KindMissing>,
    /// What the requested operation did.
    pub outcome: FieldOutcome,
    /// The field's complete option list observed before any mutation; empty when the field
    /// was not there.
    pub existing: Vec<StatusOption>,
}

/// The `Status` option names one item kind's `status_mapping` names that the field lacked.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, schemars::JsonSchema)]
pub struct KindMissing {
    /// The kind these names are configured for.
    pub kind: ItemKind,
    /// The names that kind maps a category to and the field lacked, in category order.
    // llmlint: ignore[invalid_states_unrepresentable] Each value originates from a validated
    // mapping name, as `FieldReport::missing`'s do, and the serialized string is the report's
    // intentionally simple public contract.
    pub missing: Vec<String>,
}

/// The plan and verified outcome of setting up every field a source's configuration names.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, schemars::JsonSchema)]
pub struct FieldsReport {
    /// The configured source name.
    pub source: SourceName,
    /// `Status`, always, and `Priority` when the source sets `priority_mapping`.
    // llmlint: ignore[invalid_states_unrepresentable] A list is the report's wire shape as its
    // consumer's contract fixes it — `{"source", "fields": [...]}` — so a struct with one member
    // per field would change a published JSON shape. The states the list could hold and the
    // contract forbids cannot be built: `GitHubProjectsSource::fields` is the one constructor,
    // and it pushes `Status` first and exactly once, then `Priority` exactly when configured.
    pub fields: Vec<FieldReport>,
    /// Each board text field the source's `metadata_fields` projects a value onto, in
    /// configured order. Empty — and left out of the JSON — when it projects none.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    // Kept in the schema as `"default": []` although the JSON leaves an empty list out, so
    // both SDKs model an absent list as an empty one rather than as `null`.
    #[schemars(!skip_serializing_if)]
    pub metadata_fields: Vec<MetadataFieldReport>,
}

/// The type of a board field that is not a text field, as GitHub names it — its `dataType`, or
/// its GraphQL type when it has none — which is what a projected field's conflict reports.
///
/// Never empty, never beginning with whitespace, and never `TEXT`: a text field of that name is
/// no conflict.
// The schema states that rule as `NON_TEXT_FIELD_TYPE_PATTERN`, so a generated SDK model
// refuses exactly what `try_from` below does.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, schemars::JsonSchema)]
#[serde(transparent)]
pub struct NonTextFieldType(#[schemars(regex(pattern = NON_TEXT_FIELD_TYPE_PATTERN))] String);

/// What a [`NonTextFieldType`] may be, as the emitted schema states it: a first character that
/// is not whitespace, and anything but exactly `TEXT` — spelled without a lookahead, which
/// neither SDK's validator can be relied on to support.
const NON_TEXT_FIELD_TYPE_PATTERN: &str =
    r"^(?:[^T\s]|T(?:[^E]|$)|TE(?:[^X]|$)|TEX(?:[^T]|$)|TEXT[\s\S])";

impl TryFrom<String> for NonTextFieldType {
    type Error = String;

    fn try_from(kind: String) -> Result<Self, Self::Error> {
        if kind.is_empty() || kind.starts_with(char::is_whitespace) {
            return Err("a board field's type cannot be blank or begin with whitespace".to_owned());
        }
        if kind == "TEXT" {
            return Err("a TEXT field is no conflict for a projected text field".to_owned());
        }
        Ok(Self(kind))
    }
}

impl std::fmt::Display for NonTextFieldType {
    fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        formatter.write_str(&self.0)
    }
}

/// One projected metadata text field's plan, or its verified outcome.
///
/// The setup creates a missing one as a text field and never alters a field that is there:
/// a same-named field of another type is reported in `conflict` and left as it is.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, schemars::JsonSchema)]
pub struct MetadataFieldReport {
    /// The board field's name.
    pub field: String, // llmlint: ignore[invalid_states_unrepresentable] Copied from a configuration entry `MetadataField::resolve` already validated; the serialized string is the report's intentionally simple public contract, as `FieldReport::missing`'s names are.
    /// The metadata key the value is read from.
    pub key: String, // llmlint: ignore[invalid_states_unrepresentable] As `field`: a validated configuration value reported back as written.
    /// The object keys walked inside that key's value; empty for the key's own value.
    pub path: Vec<String>,
    /// Whether the board had a text field of that name before the operation.
    // llmlint: ignore[invalid_states_unrepresentable] `exists` beside `outcome` mirrors `FieldReport`'s published shape; `GitHubProjectsSource::fields` is the one constructor and derives `outcome` from `exists` and `conflict` in one match.
    pub exists: bool,
    /// The type of a field of that name that is not a text field, which the setup leaves as
    /// it is; absent — and left out of the JSON — when there is none.
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub conflict: Option<NonTextFieldType>,
    /// What the requested operation did: `planned`; `created` for a field that was not
    /// there; `unchanged` for one that was, and for a conflict, which is never changed.
    pub outcome: FieldOutcome,
}

/// Which options one field is configured with, in the order a new field would list them.
struct FieldPlan {
    field: BoardField,
    wanted: Vec<String>,
}

/// One single-select field as the guarded setup snapshots it.
#[derive(Debug, Clone, PartialEq, Eq)]
struct SnapshotField {
    // llmlint: ignore[invalid_states_unrepresentable] This private opaque GraphQL ID is
    // passed back as the mutation's field identity; a newtype could enforce no stronger
    // invariant because GitHub publishes no grammar for it.
    field_id: String,
    options: Vec<StatusOption>,
}

/// Every single-select field of a board and every item's value of each.
#[derive(Debug, Clone, PartialEq, Eq)]
struct BoardSnapshot {
    // llmlint: ignore[invalid_states_unrepresentable] This private opaque GraphQL ID is
    // passed back as the mutation's project identity; a newtype could enforce no stronger
    // invariant because GitHub publishes no grammar for it.
    board_id: String,
    fields: BTreeMap<BoardField, SnapshotField>,
    /// Each board item's id, and its value of each field this setup owns that it holds one of.
    items: Vec<(String, BTreeMap<BoardField, AssignedStatusOption>)>,
}

impl BoardSnapshot {
    /// Every item's value of `field`, in board order — the recovery data a drift refusal
    /// carries.
    fn assignments(&self, field: BoardField) -> Vec<StatusAssignment> {
        self.items
            .iter()
            .map(|(item_id, values)| StatusAssignment {
                item_id: item_id.clone(),
                option: values.get(&field).cloned(),
            })
            .collect()
    }
}

impl GitHubProjectsSource {
    /// `content` as this source sends it, with the markup the engine handed it applied.
    fn marked(&self, content: &str) -> String {
        match self.body_markup.get() {
            Some(markup) if !content.is_empty() => markup.written(content),
            _ => content.to_owned(),
        }
    }

    /// The visible body and the metadata slot of `body` as [`metadata_body`] splits them,
    /// the visible body without the markup [`Self::marked`] applies.
    fn visible(
        &self,
        body: Option<String>,
    ) -> Result<(Option<String>, BTreeMap<String, Value>), SourceError> {
        let (visible, slot) = metadata_body(body)?;
        Ok((
            match (visible, self.body_markup.get()) {
                (Some(visible), Some(markup)) => {
                    Some(markup.read(&visible)).filter(|read| !read.is_empty())
                }
                (visible, _) => visible,
            },
            slot,
        ))
    }

    /// Report missing configured Status options and, when `apply` is true, add them with
    /// a whole-list mutation that preserves every existing id and verifies the result.
    ///
    /// # Errors
    ///
    /// Refuses a board without a single-select `Status` field. A post-write difference in
    /// any pre-existing option id or item assignment is refused with the complete pre-write
    /// assignment snapshot in the diagnostic for recovery.
    // llmlint: ignore[changed_behavior_has_e2e] The CLI journeys cover plan, no-op apply,
    // successful mutation, both drift refusals, source selection, missing Status, casing,
    // and paging. Transport errors remain the shared `graphql` boundary's behavior rather
    // than a new status-options behavior, and the pinned-schema test prevents valid GitHub
    // responses from entering the defensive malformed-response branches below.
    pub async fn status_options(
        &self,
        mode: StatusOptionsMode,
    ) -> Result<StatusOptionsReport, SourceError> {
        let before = self.status_snapshot().await?;
        // A terminal category's option is as configured as an open one's: a terminal
        // write validates it before closing and refuses when the board lacks it. Both
        // kinds' names are options of the one field, so both are asked for.
        let missing = self
            .statuses
            .wanted()
            .into_iter()
            .filter(|wanted| {
                !before
                    .options
                    .iter()
                    .any(|option| option.name.as_str().eq_ignore_ascii_case(wanted))
            })
            .collect::<Vec<_>>();
        let report = StatusOptionsReport {
            source: self.name.clone(),
            missing: missing.clone(),
            outcome: match (mode, missing.is_empty()) {
                (StatusOptionsMode::Plan, _) => StatusOptionsOutcome::Planned,
                (StatusOptionsMode::Apply, true) => StatusOptionsOutcome::Unchanged,
                (StatusOptionsMode::Apply, false) => StatusOptionsOutcome::Applied,
            },
            existing: before.options.clone(),
        };
        if mode == StatusOptionsMode::Plan || missing.is_empty() {
            return Ok(report);
        }
        let mut options = before
            .options
            .iter()
            .map(|option| {
                json!({
                    "id": option.id, "name": option.name, "color": option.color,
                    "description": option.description,
                })
            })
            .collect::<Vec<_>>();
        options.extend(missing.iter().map(|name| {
            json!({
                "name": name, "color": "GRAY", "description": ""
            })
        }));
        self.graphql(
            graphql::STATUS_OPTIONS_UPDATE,
            json!({"input": {
                "fieldId": before.field_id,
                "singleSelectOptions": options,
            }}),
        )
        .await?;
        let after = self.status_snapshot().await?;
        let options_preserved = before
            .options
            .iter()
            .all(|old| after.options.iter().any(|new| new == old));
        let additions_present = missing.iter().all(|wanted| {
            after
                .options
                .iter()
                .any(|option| option.name.as_str().eq_ignore_ascii_case(wanted))
        });
        if !options_preserved || !additions_present || after.assignments != before.assignments {
            let recovery = serde_json::to_string_pretty(&before.assignments).map_err(|error| {
                SourceError::Malformed {
                    message: format!("cannot render pre-write Status recovery snapshot: {error}"),
                }
            })?;
            return Err(SourceError::Refused {
                message: format!(
                    "GitHub changed a pre-existing Status option id, name, color or description, or an item assignment after the guarded update; the pre-write item assignment snapshot is:\n{recovery}"
                ),
            });
        }
        Ok(report)
    }

    /// A fresh snapshot of the Status field and every board item's assignment of it.
    ///
    /// # Errors
    ///
    /// Refuses a board without a single-select `Status` field, and one the token cannot see.
    async fn status_snapshot(&self) -> Result<StatusSnapshot, SourceError> {
        // Status alone, as this operation has always read it: a `Priority` field is another
        // operation's, so nothing about it can refuse this one.
        let mut board = self.board_snapshot(&[BoardField::Status]).await?;
        let field = board
            .fields
            .remove(&BoardField::Status)
            .ok_or_else(|| self.no_status_field())?;
        Ok(StatusSnapshot {
            assignments: board.assignments(BoardField::Status),
            board_id: board.board_id,
            field_id: field.field_id,
            options: field.options,
        })
    }

    /// The refusal a board with no `Status` field is answered with by the guarded setup.
    fn no_status_field(&self) -> SourceError {
        SourceError::Refused {
            message: format!("source {} board has no Status field", self.name),
        }
    }

    // llmlint: ignore-block[changed_behavior_has_e2e] Valid snapshot shapes are exercised through
    // the real CLI loopback journey, including pagination. The individual malformed guards
    // are defensive validation of a schema-pinned third-party response, not separate user
    // journeys; drift and missing-field failures cover the operation's recovery behavior.
    /// A fresh snapshot of each of the `owned` fields on the board, with its options, and of
    /// every board item's value of each, walked to the end of the board's items. A field not
    /// in `owned` is read past whatever it holds.
    async fn board_snapshot(&self, owned: &[BoardField]) -> Result<BoardSnapshot, SourceError> {
        let mut after: Option<String> = None;
        let mut snapshot: Option<BoardSnapshot> = None;
        loop {
            let data = self
                .graphql(
                    graphql::STATUS_OPTIONS_SNAPSHOT,
                    json!({
                        "owner": self.owner, "number": self.project_number,
                        "first": MAX_PAGE_SIZE, "after": after, "nestedFirst": MAX_PAGE_SIZE,
                    }),
                )
                .await?;
            let board = data
                .pointer("/owner/projectV2")
                .filter(|board| board.is_object())
                .ok_or_else(|| SourceError::Refused {
                    message: format!(
                        "source {} has no accessible GitHub Projects board",
                        self.name
                    ),
                })?;
            if board
                .pointer("/fields/pageInfo/hasNextPage")
                .and_then(Value::as_bool)
                != Some(false)
            {
                return Err(SourceError::Malformed {
                    message:
                        "GitHub project fields is incomplete or has malformed pageInfo.hasNextPage"
                            .into(),
                });
            }
            let mut fields = BTreeMap::new();
            // Only the fields this setup owns, by name: a node the single-select fragment did not
            // match carries no name, and a person's own single-select field — a `Size`, a
            // `Team` — is none of this setup's business, so nothing about it can refuse one. A
            // `Status` or `Priority` field without its options is malformed, not absent.
            // llmlint: ignore[boundary_inputs_validated] The field page this loop reads is validated as complete immediately above: any `fields.pageInfo.hasNextPage` other than `false` is refused as malformed before a node is read, so an incomplete page is never taken for the board's whole field set.
            for (owned, field) in board
                .pointer("/fields/nodes")
                .and_then(Value::as_array)
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub project fields.nodes is not an array".into(),
                })?
                .iter()
                .filter_map(|field| {
                    let named = BoardField::named(field.get("name")?.as_str()?)?;
                    owned.contains(&named).then_some((named, field))
                })
            {
                let options = field
                    .get("options")
                    .and_then(Value::as_array)
                    .ok_or_else(|| SourceError::Malformed {
                        message: "GitHub single-select field options is not an array".into(),
                    })?
                    .iter()
                    .map(|option| {
                        Ok(StatusOption {
                            id: StatusOptionId::try_from(required_str(option, "id")?.to_owned())
                                .map_err(|message| SourceError::Malformed { message })?,
                            name: ColumnName::try_from(required_str(option, "name")?.to_owned())
                                .map_err(|message| SourceError::Malformed {
                                    message: format!(
                                        "GitHub single-select option name is invalid: {message}"
                                    ),
                                })?,
                            color: serde_json::from_value(
                                option.get("color").cloned().unwrap_or(Value::Null),
                            )
                            .map_err(|error| {
                                SourceError::Malformed {
                                    message: format!(
                                        "GitHub single-select option color is invalid: {error}"
                                    ),
                                }
                            })?,
                            description: optional_str(option, "description")?
                                .unwrap_or_default()
                                .to_owned(),
                        })
                    })
                    .collect::<Result<Vec<_>, SourceError>>()?;
                let snapshot = SnapshotField {
                    field_id: required_nonblank_str(field, "id")?.to_owned(),
                    options,
                };
                // A board's field names are unique, so a second one is an answer that cannot
                // say which field the setup would act on — refused rather than one chosen.
                if fields.insert(owned, snapshot).is_some() {
                    return Err(SourceError::Malformed {
                        message: format!(
                            "GitHub answered two {} fields for this board",
                            owned.name()
                        ),
                    });
                }
            }
            let board_id = required_nonblank_str(board, "id")?.to_owned();
            let current = snapshot.get_or_insert_with(|| BoardSnapshot {
                board_id,
                fields,
                items: Vec::new(),
            });
            let items = board
                .pointer("/items/nodes")
                .and_then(Value::as_array)
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub project items.nodes is not an array".into(),
                })?;
            for item in items {
                let field_values =
                    item.get("fieldValues")
                        .ok_or_else(|| SourceError::Malformed {
                            message: "GitHub project item is missing fieldValues".into(),
                        })?;
                if field_values
                    .pointer("/pageInfo/hasNextPage")
                    .and_then(Value::as_bool)
                    != Some(false)
                {
                    return Err(SourceError::Malformed {
                        message: "GitHub project item fieldValues is incomplete or has malformed pageInfo.hasNextPage".into(),
                    });
                }
                let values = item
                    .pointer("/fieldValues/nodes")
                    .and_then(Value::as_array)
                    .ok_or_else(|| SourceError::Malformed {
                        message: "GitHub project item fieldValues.nodes is not an array".into(),
                    })?;
                let item_id = required_nonblank_str(item, "id")?;
                let mut assigned = BTreeMap::new();
                for value in values {
                    let Some(field) = value
                        .pointer("/field/name")
                        .and_then(Value::as_str)
                        .and_then(BoardField::named)
                        .filter(|field| owned.contains(field))
                    else {
                        continue;
                    };
                    let held = assigned.insert(
                        field,
                        AssignedStatusOption {
                            id: StatusOptionId::try_from(
                                required_str(value, "optionId")?.to_owned(),
                            )
                            .map_err(|message| SourceError::Malformed { message })?,
                            name: ColumnName::try_from(required_str(value, "name")?.to_owned())
                                .map_err(|message| SourceError::Malformed {
                                    message: format!(
                                        "GitHub assigned {} name is invalid: {message}",
                                        field.name()
                                    ),
                                })?,
                        },
                    );
                    // An item holds one value of a field, so a second one leaves no way to
                    // tell which it holds — and a verification or recovery built on either
                    // could restore the wrong one.
                    if held.is_some() {
                        return Err(SourceError::Malformed {
                            message: format!(
                                "GitHub answered two {} values for board item {item_id}",
                                field.name()
                            ),
                        });
                    }
                }
                current.items.push((item_id.to_owned(), assigned));
            }
            let page = board.get("items").ok_or_else(|| SourceError::Malformed {
                message: "GitHub project is missing items".into(),
            })?;
            let has_next = page
                .pointer("/pageInfo/hasNextPage")
                .and_then(Value::as_bool)
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub project items.pageInfo.hasNextPage is not a boolean".into(),
                })?;
            if !has_next {
                break;
            }
            let next =
                required_nonblank_str(page.get("pageInfo").unwrap_or(&Value::Null), "endCursor")?;
            validate_cursor_progress(after.as_deref(), next)?;
            after = Some(next.to_owned());
        }
        snapshot.ok_or_else(|| SourceError::Malformed {
            message: "GitHub returned no board field snapshot".into(),
        })
    }
    // llmlint: ignore-end[changed_behavior_has_e2e]

    /// Report every board field this source's configuration names and, with
    /// [`SetupMode::Apply`], set each up: add the options a field lacks, and create
    /// the `Priority` field when the board has none.
    ///
    /// The fields are `Status`, always, with the options `status_mapping` resolves to; and
    /// `Priority`, when `priority_mapping` is set, with its four mapped options — created in
    /// the order urgent, high, medium, low. An option a field already has keeps its id, name,
    /// color and description: the whole option list goes back with every existing id, because
    /// a re-minted id clears every item's value.
    ///
    /// # Errors
    ///
    /// Refuses a board without a single-select `Status` field. After an apply the board is
    /// read again, and a pre-existing option or any item's value of either field that moved is
    /// refused with the complete pre-write assignments in the diagnostic, for recovery.
    // llmlint: ignore[changed_behavior_has_e2e] The `sources fields` journeys drive plan,
    // unchanged apply, a created field, an added option to each field, drift refusal, a board
    // with no Status field and a non-github-projects source through the compiled CLI against
    // the loopback board. Transport errors are the shared `graphql` boundary's behavior.
    pub async fn fields(&self, mode: SetupMode) -> Result<FieldsReport, SourceError> {
        let owned: Vec<BoardField> = if self.priorities.is_some() {
            vec![BoardField::Status, BoardField::Priority]
        } else {
            vec![BoardField::Status]
        };
        let before = self.board_snapshot(&owned).await?;
        let mut plans = vec![FieldPlan {
            field: BoardField::Status,
            wanted: self.statuses.wanted(),
        }];
        if !before.fields.contains_key(&BoardField::Status) {
            return Err(self.no_status_field());
        }
        if let Some(mapping) = &self.priorities {
            plans.push(FieldPlan {
                field: BoardField::Priority,
                wanted: mapping.names().map(str::to_owned).collect(),
            });
        }
        // The snapshot reads single-select fields alone, so a field it did not find may still
        // be on the board under the name, of another type: creating one beside it would fail
        // part way, or leave two fields of one name. Asked of the board's own field list, and
        // only when a field is missing.
        if plans
            .iter()
            .any(|plan| !before.fields.contains_key(&plan.field))
        {
            let board = self.board_fields().await?;
            for plan in plans
                .iter()
                .filter(|plan| !before.fields.contains_key(&plan.field))
            {
                if let Some(field) = Board::field(&board.fields, plan.field.name())? {
                    return Err(SourceError::Refused {
                        message: format!(
                            "source {}'s board has a {} field that is not a single-select field \
                             (it is a {}), so it cannot hold this source's options; next: rename \
                             or remove that field, then run this again",
                            self.name,
                            plan.field.name(),
                            optional_str(field, "__typename")?.unwrap_or("field of another type")
                        ),
                    });
                }
            }
        }
        let mut reports = Vec::new();
        for plan in &plans {
            let held = before.fields.get(&plan.field);
            let existing = held.map(|field| field.options.clone()).unwrap_or_default();
            let mut missing: Vec<String> = Vec::new();
            for wanted in &plan.wanted {
                let present = existing
                    .iter()
                    .any(|option| option.name.as_str().eq_ignore_ascii_case(wanted))
                    || missing
                        .iter()
                        .any(|named| named.eq_ignore_ascii_case(wanted));
                if !present {
                    missing.push(wanted.clone());
                }
            }
            let kinds = match plan.field {
                BoardField::Status => self.statuses.missing_by_kind(&existing),
                BoardField::Priority => Vec::new(),
            };
            reports.push(FieldReport {
                field: plan.field,
                exists: held.is_some(),
                kinds,
                outcome: match (mode, held.is_some(), missing.is_empty()) {
                    (SetupMode::Plan, _, _) => FieldOutcome::Planned,
                    (SetupMode::Apply, true, true) => FieldOutcome::Unchanged,
                    (SetupMode::Apply, true, false) => FieldOutcome::Applied,
                    (SetupMode::Apply, false, _) => FieldOutcome::Created,
                },
                missing,
                existing,
            });
        }
        // A projected field is a text field, which the single-select snapshot cannot see, so
        // it is asked of the board's own field list — and only when one is configured.
        let mut metadata_fields = Vec::new();
        if !self.metadata_fields.is_empty() {
            let board = self.board_fields().await?;
            for projection in &self.metadata_fields {
                let conflict = match Board::field(&board.fields, &projection.field)? {
                    None => None,
                    Some(field) => {
                        let typename = optional_str(field, "__typename")?;
                        let data_type = optional_str(field, "dataType")?;
                        if typename == Some("ProjectV2Field") && data_type == Some("TEXT") {
                            None
                        } else {
                            let named = data_type
                                .filter(|kind| *kind != "TEXT")
                                .or(typename)
                                .unwrap_or("field of another type");
                            Some(NonTextFieldType::try_from(named.to_owned()).map_err(
                                |message| SourceError::Malformed {
                                    message: format!(
                                        "GitHub answered the board's {} field with a type this \
                                         setup cannot read: {message}",
                                        &*projection.field
                                    ),
                                },
                            )?)
                        }
                    }
                };
                let exists =
                    conflict.is_none() && Board::field(&board.fields, &projection.field)?.is_some();
                metadata_fields.push(MetadataFieldReport {
                    field: projection.field.to_string(),
                    key: projection.key.to_string(),
                    path: projection.path.clone(),
                    exists,
                    outcome: match (mode, exists || conflict.is_some()) {
                        (SetupMode::Plan, _) => FieldOutcome::Planned,
                        (SetupMode::Apply, true) => FieldOutcome::Unchanged,
                        (SetupMode::Apply, false) => FieldOutcome::Created,
                    },
                    conflict,
                });
            }
        }
        let report = FieldsReport {
            source: self.name.clone(),
            fields: reports,
            metadata_fields,
        };
        let writes: Vec<&FieldReport> = report
            .fields
            .iter()
            .filter(|field| !field.missing.is_empty() || !field.exists)
            .collect();
        let creates: Vec<&MetadataFieldReport> = report
            .metadata_fields
            .iter()
            .filter(|field| field.outcome == FieldOutcome::Created)
            .collect();
        if mode == SetupMode::Plan || (writes.is_empty() && creates.is_empty()) {
            return Ok(report);
        }
        let mut landed: Vec<&str> = Vec::new();
        for field in &writes {
            let added = field
                .missing
                .iter()
                .map(|name| json!({"name": name, "color": "GRAY", "description": ""}));
            let sent = match before.fields.get(&field.field) {
                Some(held) => {
                    let mut options = held
                        .options
                        .iter()
                        .map(|option| {
                            json!({
                                "id": option.id, "name": option.name, "color": option.color,
                                "description": option.description,
                            })
                        })
                        .collect::<Vec<_>>();
                    options.extend(added);
                    self.graphql(
                        graphql::STATUS_OPTIONS_UPDATE,
                        json!({"input": {
                            "fieldId": held.field_id,
                            "singleSelectOptions": options,
                        }}),
                    )
                    .await
                }
                None => {
                    self.graphql(
                        graphql::CREATE_FIELD,
                        json!({"input": {
                            "projectId": before.board_id, "dataType": "SINGLE_SELECT",
                            "name": field.field.name(),
                            "singleSelectOptions": added.collect::<Vec<_>>(),
                        }}),
                    )
                    .await
                }
            };
            // A mutation that failed does not establish that GitHub left its field as it was,
            // so every failure from here on carries the recovery data a drift refusal does.
            match sent {
                Ok(_) => landed.push(field.field.name()),
                Err(error) => {
                    let changed = if landed.is_empty() {
                        String::new()
                    } else {
                        format!("changed the {} field and then ", landed.join(" and "))
                    };
                    return Err(SourceError::Refused {
                        message: format!(
                            "the guarded field setup {changed}failed on the {} field, which it may \
                             have changed part way: {error}; the pre-write item assignments \
                             are:\n{}",
                            field.field.name(),
                            recovery(&report, &before)?
                        ),
                    });
                }
            }
        }
        for field in &creates {
            let sent = self
                .graphql(
                    graphql::CREATE_FIELD,
                    json!({"input": {
                        "projectId": before.board_id, "dataType": "TEXT", "name": field.field,
                    }}),
                )
                .await;
            if let Err(error) = sent {
                let changed = if landed.is_empty() {
                    String::new()
                } else {
                    format!("changed the {} field and then ", landed.join(" and "))
                };
                return Err(SourceError::Refused {
                    message: format!(
                        "the guarded field setup {changed}failed creating the {} text field: \
                         {error}; run it again to create what is still missing",
                        field.field
                    ),
                });
            }
            landed.push(&field.field);
        }
        // The board has been written, so a verification read that fails leaves it unverified
        // rather than unchanged, and says what to put back.
        let after = match self.board_snapshot(&owned).await {
            Ok(after) => after,
            Err(error) => {
                return Err(SourceError::Refused {
                    message: format!(
                        "the guarded field setup changed the {} field and then could not read the \
                         board back to verify it: {error}; the pre-write item assignments are:\n{}",
                        landed.join(" and "),
                        recovery(&report, &before)?
                    ),
                });
            }
        };
        let mut moved = Vec::new();
        for field in &report.fields {
            let name = field.field.name();
            let now = after
                .fields
                .get(&field.field)
                .map(|held| held.options.as_slice())
                .unwrap_or_default();
            if !field.existing.iter().all(|old| now.contains(old)) {
                moved.push(format!(
                    "a pre-existing {name} option id, name, color or description"
                ));
            }
            if !field.missing.iter().all(|wanted| {
                now.iter()
                    .any(|option| option.name.as_str().eq_ignore_ascii_case(wanted))
            }) {
                moved.push(format!("an added {name} option"));
            }
            if after.assignments(field.field) != before.assignments(field.field) {
                moved.push(format!("an item's {name} value"));
            }
        }
        if !creates.is_empty() {
            // Read afresh: the view held for this command is the one from before the creates.
            *self.fields_cache()? = None;
            let board = match self.board_fields().await {
                Ok(board) => board,
                Err(error) => {
                    return Err(SourceError::Refused {
                        message: format!(
                            "the guarded field setup changed the {} field and then could not \
                             read the board back to verify it: {error}; run it again to verify \
                             it, which creates nothing a second time",
                            landed.join(" and ")
                        ),
                    });
                }
            };
            for field in &creates {
                let text = Board::field(&board.fields, &field.field)?.is_some_and(|held| {
                    held.get("__typename").and_then(Value::as_str) == Some("ProjectV2Field")
                        && held.get("dataType").and_then(Value::as_str) == Some("TEXT")
                });
                if !text {
                    moved.push(format!("the created {} text field", field.field));
                }
            }
        }
        if !moved.is_empty() {
            return Err(SourceError::Refused {
                message: format!(
                    "GitHub changed {} after the guarded field setup; the pre-write item \
                     assignments are:\n{}",
                    moved.join(", "),
                    recovery(&report, &before)?
                ),
            });
        }
        Ok(report)
    }

    /// Validate configuration and capture the named credential without exposing it.
    ///
    /// # Errors
    ///
    /// Returns [`SourceError::Config`] for a configuration this instance cannot use and
    /// [`SourceError::Auth`] when the named credential is missing or empty.
    pub fn new(
        name: &SourceName,
        config: GitHubProjectsConfig,
        secrets: &dyn SecretResolver,
    ) -> Result<Self, SourceError> {
        Self::recording_into(name, config, secrets, Arc::new(Accounting::new()))
    }

    /// The same, recording every request it sends into an accounting the caller holds too.
    ///
    /// [`Self::new`] is this with an accounting of its own. A caller that is also making
    /// its own calls to GitHub — a lane verifying a schema, sweeping residue or cleaning
    /// up — passes the one it records those into, so the session total accounts for the
    /// whole session rather than for this source's share of it.
    ///
    /// # Errors
    ///
    /// Exactly [`Self::new`]'s: [`SourceError::Config`] for a configuration this instance
    /// cannot use and [`SourceError::Auth`] when the named credential is missing or empty.
    pub fn recording_into(
        name: &SourceName,
        config: GitHubProjectsConfig,
        secrets: &dyn SecretResolver,
        ledger: Arc<Accounting>,
    ) -> Result<Self, SourceError> {
        if !valid_github_owner(&config.owner) {
            return Err(SourceError::Config {
                message: "owner must be 1-39 ASCII letters, digits, or single hyphens, and cannot start or end with a hyphen".into(),
            });
        }
        if config.project_number == 0 || config.project_number > i32::MAX as u32 {
            return Err(SourceError::Config {
                message: format!("project_number must be between 1 and {}", i32::MAX),
            });
        }
        if !valid_environment_name(&config.token_env) {
            return Err(SourceError::Config {
                message: "token_env must be a valid environment-variable name".into(),
            });
        }
        let repository = config
            .repository
            .as_deref()
            .map(RepositoryTarget::parse)
            .transpose()?;
        if let Some(entry) = config
            .creation_owners
            .iter()
            .find(|entry| !valid_github_owner(entry))
        {
            return Err(SourceError::Config {
                message: format!(
                    "creation_owners entry {entry:?} is not a GitHub login: a login is 1-39 ASCII \
                     letters, digits, or single hyphens, and cannot start or end with a hyphen"
                ),
            });
        }
        let endpoint = Url::parse(&config.endpoint).map_err(|e| SourceError::Config {
            message: format!("endpoint is not a valid URL: {e}"),
        })?;
        if endpoint.scheme() != "https"
            && !(endpoint.scheme() == "http"
                && endpoint
                    .host_str()
                    .is_some_and(|h| h == "127.0.0.1" || h == "localhost" || h == "::1"))
        {
            return Err(SourceError::Config {
                message:
                    "endpoint must use HTTPS (HTTP is accepted only for a loopback test server)"
                        .into(),
            });
        }
        let token = secrets.get(&config.token_env).filter(|token| !token.expose_secret().trim().is_empty()).ok_or_else(|| SourceError::Auth {
            message: format!("environment variable {} is missing or empty; set it to a fine-grained GitHub token granting Projects and Issues read/write plus Pull requests read-only access for every repository represented on the board", config.token_env),
        })?;
        Ok(Self {
            name: name.clone(),
            owner: config.owner,
            project_number: config.project_number,
            repository,
            creation_owners: config.creation_owners,
            asset_client: std::sync::OnceLock::new(),
            body_markup: std::sync::OnceLock::new(),
            endpoint,
            token,
            credential_name: config.token_env,
            statuses: BoardStatuses::resolve(&config.status_mapping, name)?,
            priorities: config
                .priority_mapping
                .map(|mapping| PriorityMapping::resolve(mapping, name))
                .transpose()?,
            metadata_fields: MetadataField::resolve(config.metadata_fields, name)?,
            sub_issues: config.sub_issues,
            assignees: config.assignees,
            people: Mutex::new(BTreeMap::new()),
            forgotten: Mutex::new(Vec::new()),
            spellings: Mutex::new(BTreeMap::new()),
            link_types: onetaskgraph_plugin_api::LinkTypes::checked(
                KIND,
                &config.link_types,
                &[relations::RELATES_TO],
                relations::REFUSED_WORDS,
            )
            .map_err(|error| match error {
                SourceError::Config { message } => SourceError::Config {
                    message: format!("source {name}: {message}"),
                },
                other => other,
            })?,
            client: std::sync::OnceLock::new(),
            created: Mutex::new(Vec::new()),
            updated: Mutex::new(Vec::new()),
            commented: Mutex::new(Vec::new()),
            pacing: Pacing::resolve(config.pacing, name)?,
            last_mutation: Mutex::new(None),
            last_served: Mutex::new(None),
            windows: Mutex::new(pacing::Windows::default()),
            task_children: Mutex::new(BTreeMap::new()),
            written: Mutex::new(BTreeMap::new()),
            deferred: Mutex::new(None),
            held_updates: Mutex::new(Vec::new()),
            held_comments: Mutex::new(BTreeMap::new()),
            task_writes: Mutex::new(None),
            written_tokens: Mutex::new(BTreeMap::new()),
            clock: system_clock(),
            declared_private: std::sync::atomic::AtomicBool::new(false),
            numeric_repositories: tokio::sync::Mutex::new(BTreeMap::new()),
            board_cache: Mutex::new(None),
            search_cache: Mutex::new(None),
            narrowed_cache: Mutex::new(BTreeMap::new()),
            resolved_cache: Mutex::new(BTreeMap::new()),
            read_ahead: Mutex::new(BTreeMap::new()),
            pending_origins: Mutex::new(Vec::new()),
            member_classes: Mutex::new(BTreeMap::new()),
            carried: Mutex::new(BTreeMap::new()),
            held_cache: Mutex::new(BTreeMap::new()),
            children_cache: Mutex::new(BTreeMap::new()),
            search_next: Mutex::new(BTreeMap::new()),
            fields_cache: Mutex::new(None),
            repository_cache: Mutex::new(BTreeMap::new()),
            ledger,
            expected: expected::Held::default(),
            projection: projection::Held::default(),
        })
    }

    /// The client every API request goes through, built on the first one.
    ///
    /// Not with the source: building a client loads the system's certificate store, which
    /// costs tens of milliseconds, and a command that configures this board but never sends to
    /// it — a read of another source — should not pay that in its process.
    fn client(&self) -> Result<&Client, SourceError> {
        built(&self.client, || {
            Client::builder()
                .user_agent("onetaskgraph")
                .build()
                .map_err(|e| SourceError::Config {
                    message: format!("cannot build HTTP client: {e}"),
                })
        })
    }

    /// The client attachment requests go through, with its own redirect policy, built on the
    /// first one for the reason [`Self::client`] gives.
    fn asset_client(&self) -> Result<&Client, SourceError> {
        built(&self.asset_client, || assets::client(&self.endpoint))
    }

    /// A snapshot of every request this source has sent, and what each cost.
    ///
    /// A value to hold and compare rather than a borrow of the accounting itself, so two
    /// of them can sit side by side. When this source was built with
    /// [`Self::recording_into`] the snapshot is the whole shared session, which is the
    /// point of building it that way.
    #[must_use]
    pub fn accounting(&self) -> accounting::Session {
        self.ledger.snapshot()
    }

    /// Send one GraphQL document, pacing this source's own mutations and waiting out a
    /// rate limit rather than handing it straight back as an error.
    ///
    /// Retrying is safe for every document here, including the mutations, and the reason
    /// is that only a *refusal* is retried: [`Limiter::classify`] rules on a response
    /// GitHub sent, and a request GitHub refused for a rate limit did not run, so nothing
    /// this replays has already taken effect. An outcome this source cannot know — the
    /// send failed, or the body could not be read, so the mutation may well have landed —
    /// is [`Attempt::Failed`] in [`send_once`] and leaves this loop without a second
    /// attempt. A duplicate write would come from replaying one of those, and none is
    /// replayed.
    async fn graphql(&self, query: &str, variables: Value) -> Result<Value, SourceError> {
        // Updates held back go out before any other request, so whatever it reads or writes
        // sees the board as though each had been sent when it was made.
        Box::pin(self.send_held_updates()).await?;
        // Origins held back for a read ahead that never came are sent before anything else.
        if query != graphql::ORIGIN_LOOKUPS && query != graphql::ORIGIN_LOOKUPS_WITH_MEMBERS {
            Box::pin(self.send_pending_origins(None)).await?;
        }
        let journaling = self.expected.journaling();
        let spelling = self.spelling();
        if !journaling && !spelling {
            return self.graphql_sent(query, variables).await;
        }
        // While a write the store's cache expects runs, its mutations select the whole issue
        // they leave behind, so the response vouches for what the cache keeps — and every
        // request is recorded, so the write's parts are reported from their own responses.
        // A targeted update sends a twin of each selecting the issue's token parts inline, at
        // the same price, so the version it answers is the one its own last response spells.
        let (query, variables) = match (journaling, graphql::whole_twin(query)) {
            (true, Some(twin)) => (twin, expected::whole_variables(variables)),
            _ => match graphql::token_twin(query).filter(|_| spelling) {
                Some(twin) => (twin, token_variables(variables)),
                None => (query, variables),
            },
        };
        let sent = self.graphql_sent(query, variables.clone()).await;
        if spelling && let Ok(data) = &sent {
            self.record_written_token(query, &variables, data);
        }
        if journaling {
            self.journal(query, &variables, &sent).await;
        }
        sent
    }

    /// Whether a targeted update of this source is running, whose mutations are each sent as
    /// the twin selecting the whole issue (see [`Self::written_tokens`]).
    fn spelling(&self) -> bool {
        SPELLING
            .try_with(|source| *source == std::ptr::from_ref(self) as usize)
            .unwrap_or(false)
    }

    /// Record the change token the issue a mutation's response shows whole holds now, as
    /// [`TaskSource::written_task`] answers it; a response showing no whole issue records
    /// nothing.
    fn record_written_token(&self, document: &str, variables: &Value, data: &Value) {
        let (issue, _) = expected::read_back(document, variables, data);
        if let Some(issue) = issue {
            self.record_written_issue(issue);
        }
    }

    /// Record the change token `issue`, as a write's own answer shows it whole, holds now — or,
    /// where that answer lacks a part of one, that it was shown and spells none.
    fn record_written_issue(&self, issue: &Value) {
        // An answer naming the issue by its id alone did not show it, so it says nothing.
        if issue.get("updatedAt").is_none() {
            return;
        }
        let Some(id) = issue.get("id").and_then(Value::as_str) else {
            return;
        };
        let (item, token) = match self.written_token_of(issue) {
            Some((item, token)) => (item, Some(token)),
            None => {
                // Answered whole, but without every part of a token: its board item's id, where
                // it says one, so a later write naming that item still drops the entry.
                let item = issue
                    .pointer("/projectItems/nodes")
                    .and_then(Value::as_array)
                    .and_then(|nodes| self.board_entry(nodes))
                    .and_then(|entry| entry.get("id"))
                    .and_then(Value::as_str)
                    .unwrap_or_default()
                    .to_owned();
                (item, None)
            }
        };
        if let Ok(mut tokens) = self.written_tokens.lock() {
            tokens.insert(NativeId(id.to_owned()), (item, token));
        }
    }

    /// Record `token` as the change token the task `item` holds after this command's write.
    fn remember_written_token(&self, item: &Resolved, token: String) {
        if let Ok(mut tokens) = self.written_tokens.lock() {
            tokens.insert(item.id.clone(), (item.item_id.clone(), Some(token)));
        }
    }

    /// The change token the task `id` holds after this command's latest write of it, when that
    /// write's own response — or the read made right after it — showed one.
    fn written_token(&self, id: &NativeId) -> Option<String> {
        self.written_answer(id).flatten()
    }

    /// What this command's latest write of the task `id` showed of it: `Some` once a response
    /// showed the issue whole — its token, or `None` where that response could not spell one —
    /// and `None` where no response did.
    fn written_answer(&self, id: &NativeId) -> Option<Option<String>> {
        self.written_tokens
            .lock()
            .ok()?
            .get(id)
            .map(|(_, token)| token.clone())
    }

    /// Drop the written token of every task `variables` names, by its id or its board item's,
    /// before a mutation that may move it is sent.
    fn forget_written_tokens(&self, variables: &Value) {
        let Ok(mut tokens) = self.written_tokens.lock() else {
            return;
        };
        if tokens.is_empty() {
            return;
        }
        let named: Vec<&str> = variables
            .as_object()
            .into_iter()
            .flat_map(|arguments| arguments.values())
            .flat_map(|input| {
                WRITTEN_NAMES
                    .iter()
                    .filter_map(|key| input.get(key).and_then(Value::as_str))
            })
            .collect();
        tokens.retain(|id, (item, _)| {
            !named
                .iter()
                .any(|name| *name == id.0 || *name == item.as_str())
        });
    }

    /// One document sent, retried while GitHub asks for a wait, exactly as asked.
    async fn graphql_sent(&self, query: &str, variables: Value) -> Result<Value, SourceError> {
        if is_mutation(query) {
            self.forget_written_tokens(&variables);
        }
        if is_mutation(query)
            && ![
                graphql::ADD_COMMENT,
                graphql::UPDATE_COMMENT,
                graphql::DELETE_COMMENT,
                graphql::ADD_COMMENT_WHOLE,
                graphql::UPDATE_COMMENT_WHOLE,
            ]
            .contains(&query)
        {
            let mut cache = self.resolved_cache()?;
            let mut ahead = self.read_ahead()?;
            let mut carried = self.carried()?;
            for argument in ["input", "second", "third", "clear"] {
                if let Some(input) = variables.get(argument) {
                    let names = |id: &NativeId, item_id: Option<&str>| {
                        ["id", "issueId", "subjectId", "itemId"].iter().any(|key| {
                            input
                                .get(key)
                                .and_then(Value::as_str)
                                .is_some_and(|value| value == id.0 || Some(value) == item_id)
                        })
                    };
                    cache.retain(|id, item| !names(id, Some(&item.item_id)));
                    ahead.retain(|id, item| {
                        !names(id, item.as_ref().map(|item| item.item_id.as_str()))
                    });
                    carried.retain(|id, item| !names(id, Some(&item.item_id)));
                }
            }
        }
        let doing = operation_description(query);
        let mutation = is_mutation(query);
        let charge = pacing::Charge::graphql(query, &variables);
        let mut waited = Duration::ZERO;
        let mut waits = 0_u32;
        let mut backoff = self.pacing.retry_backoff;
        loop {
            let ticket = self.pace(charge, mutation).await;
            // Every write but a create is asked for immediately before it is sent — a create's
            // fields, its relationships, its content and the delete that takes it back included
            // — so a copy that owns the create renews its hold first, and one that has lost it
            // sends nothing more. Asked again before a retry, which is sent later.
            if mutation
                && !query.contains("createIssue(")
                && let Err(refused) = <dyn onetaskgraph_plugin_api::WriteGuard>::check().await
            {
                self.paced(ticket, mutation);
                return Err(refused);
            }
            let attempt = self.send_once(query, &variables, waits + 1).await;
            self.paced(ticket, mutation);
            let remaining = self.pacing.retry_budget.saturating_sub(waited);
            let (wait, cause) = match attempt {
                Ok(data) => return Ok(data),
                Err(Attempt::Failed(error)) => return Err(error),
                Err(Attempt::Conflicted(said)) => {
                    // A create is never replayed: see `CONFLICT_WORDINGS`.
                    if query.contains("createIssue(") || backoff.is_zero() || backoff > remaining {
                        return Err(SourceError::Unavailable {
                            message: format!(
                                "GitHub refused this source while {doing} as a temporary \
                                 conflict ({said}); it retried {} over {} and the conflict \
                                 did not clear within the {} one call may spend waiting. \
                                 next: run the command again once nothing else is writing to \
                                 the same items",
                                plural(waits, "time"),
                                seconds(waited),
                                seconds(self.pacing.retry_budget),
                            ),
                        });
                    }
                    (backoff, pacing::cause::BACKOFF)
                }
                Err(Attempt::Limited(limited)) => {
                    let (wait, cause) = limited.wait(backoff);
                    // A wait of nothing spends none of the budget, so it is exhaustion rather
                    // than a retry. `Pacing::resolve` rules out every way of configuring one
                    // except a budget of zero, where reporting the first refusal is the ask.
                    if wait.is_zero() || wait > remaining {
                        return Err(limited.exhausted(
                            doing,
                            waits,
                            waited,
                            wait,
                            self.pacing.retry_budget,
                        ));
                    }
                    (wait, cause)
                }
            };
            self.clock.sleep_because(wait, cause).await;
            waited += wait;
            waits += 1;
            backoff = backoff.saturating_mul(2);
        }
    }

    /// Hold one request charging `charge` until every secondary window it charges has room for
    /// the whole of it — and, for a mutation of an instance configured with
    /// `pacing.min_mutation_interval_ms`, until that interval has passed since the last one
    /// finished — then count it, answering the ticket [`Self::paced`] finishes it with.
    ///
    /// The moment is decided and counted under the lock and waited out outside it, so two
    /// callers take two places rather than the same one, and no lock is held across an await.
    /// A request with room goes out at once: nothing here waits that a window does not ask for.
    async fn pace(&self, charge: pacing::Charge, mutation: bool) -> pacing::Ticket {
        let (wait, cause, ticket) = {
            // A poisoned lock here costs pacing, not correctness, and refusing the write over
            // it would turn an earlier failure into a second one for no gain.
            let mut windows = self
                .windows
                .lock()
                .unwrap_or_else(std::sync::PoisonError::into_inner);
            let now = self.clock.now();
            let (mut at, mut cause) = if self.pacing.windows {
                windows.room(now, charge)
            } else {
                (now, None)
            };
            if mutation && !self.pacing.min_mutation_interval.is_zero() {
                let mut last = self
                    .last_mutation
                    .lock()
                    .unwrap_or_else(std::sync::PoisonError::into_inner);
                // `checked_add` rather than `+`: adding durations can panic on overflow, and
                // pacing is not worth a panic even at a bound `MAX_PACING_MS` already rules out.
                if let Some(earliest) = last
                    .and_then(|previous| previous.checked_add(self.pacing.min_mutation_interval))
                    && earliest > at
                {
                    at = earliest;
                    cause = Some(pacing::cause::MIN_INTERVAL);
                }
                *last = Some(at);
            }
            let ticket = windows.charge(at, charge);
            (at.saturating_sub(now), cause, ticket)
        };
        if let (false, Some(cause)) = (wait.is_zero(), cause) {
            self.clock.sleep_because(wait, cause).await;
        }
        ticket
    }

    /// Record that the request `ticket` was charged for has finished, so each window — and
    /// the configured spacing — counts it from here rather than from the moment it left.
    ///
    /// This source can only choose when a request *departs*; GitHub's limiter counts when it
    /// *arrives*, and the two differ by whatever the request spent in transit. Counting from
    /// departure would hand the limiter a window that looked roomier than it is, so a source
    /// pacing exactly at a limit could still be seen arriving over it — which is how a copy
    /// paced well inside a board's threshold was once refused by it on a slower machine while
    /// passing on a quick one. The previous request had arrived before its response came back,
    /// so counting from here is never earlier than the limiter does, whatever transit costs
    /// and on whatever platform: the safe side of a limit that punishes being wrong by refusing
    /// reads for the next fifty minutes.
    ///
    /// A failed attempt is finished too: a request refused by the limiter still arrived, and
    /// one that never left costs only room nobody needed.
    fn paced(&self, ticket: pacing::Ticket, mutation: bool) {
        let now = self.clock.now();
        self.windows
            .lock()
            .unwrap_or_else(std::sync::PoisonError::into_inner)
            .finish(ticket, now);
        if mutation && !self.pacing.min_mutation_interval.is_zero() {
            let mut last = self
                .last_mutation
                .lock()
                .unwrap_or_else(std::sync::PoisonError::into_inner);
            // `max` rather than an assignment: a concurrent caller may already have reserved a
            // slot further out, and completing this request must never pull that slot back in.
            *last = Some(last.map_or(now, |reserved| reserved.max(now)));
        }
    }

    /// One HTTP attempt, classified into an answer, a rate limit to wait out, or a
    /// failure that waiting cannot help — and recorded, whichever of the three it was.
    ///
    /// This is the one place a request leaves this crate, which is why the accounting is
    /// here rather than at each of the callers: a read path added later is counted without
    /// anybody remembering to count it, and
    /// `the_session_report_counts_every_request_the_board_served_and_what_each_cost` fails
    /// when one is not.
    async fn send_once(
        &self,
        query: &str,
        variables: &Value,
        number: u32,
    ) -> Result<Value, Attempt> {
        // A lookup goes out sized to the people it names, and the record names and counts it as
        // what it is; everything that decides how it is retried or read compares `query`, the
        // constant it was written as.
        let sized = graphql::sized(query, variables);
        let executed = self.ledger.priced_document(sized.as_ref());
        let Attempted {
            result,
            limits,
            reported_cost,
            response_outcome,
        } = self.attempt(executed.as_ref(), variables).await;
        // No `otherwise` name: every document this source sends is one of its own, and the
        // inventory gate on `graphql::DOCUMENTS` is what keeps that true.
        let sending = accounting::Request::graphql(sized.as_ref(), variables, None, reported_cost);
        let outcome = match &result {
            Ok(_) => accounting::Outcome::Answered,
            Err(Attempt::Limited(_)) => accounting::Outcome::RateLimited,
            Err(Attempt::Failed(_) | Attempt::Conflicted(_)) => accounting::Outcome::Refused,
        };
        self.ledger
            .record(sending.finished(outcome, limits).with_execution(
                Some(executed.into_owned()),
                response_outcome,
                number,
            ));
        result
    }

    /// The attempt itself, with what its response said about the rate limit alongside.
    ///
    /// The two are returned together rather than recorded here because every one of the
    /// early exits below is a different outcome, and a record written at each of them is a
    /// record one of them can be added without.
    async fn attempt(&self, query: &str, variables: &Value) -> Attempted {
        let mut limits = accounting::RateLimit::default();
        let mut reported_cost = None;
        let mut response_outcome = accounting::ResponseOutcome::TransportError;
        // Spelled here, at the one place a document leaves for GitHub, and nowhere earlier:
        // everything that decides how a document is retried or how its answer is read compares
        // it with the constant it was written as.
        let spelled = self
            .assignees
            .then(|| graphql::with_assignees(query))
            .flatten();
        let result = self
            .attempted(
                query,
                spelled.as_deref().unwrap_or(query),
                variables,
                &mut limits,
                &mut reported_cost,
                &mut response_outcome,
            )
            .await;
        Attempted {
            result,
            limits,
            reported_cost,
            response_outcome,
        }
    }

    /// One HTTP attempt, filling in what its response said about the rate limit as it goes.
    async fn attempted(
        &self,
        query: &str,
        sent: &str,
        variables: &Value,
        limits: &mut accounting::RateLimit,
        reported_cost: &mut Option<u64>,
        response_outcome: &mut accounting::ResponseOutcome,
    ) -> Result<Value, Attempt> {
        let response = self
            .client()
            .map_err(Attempt::Failed)?
            .post(self.endpoint.clone())
            .bearer_auth(self.token.expose_secret())
            .json(&json!({"query": sent, "variables": variables}))
            .send()
            .await
            .map_err(|e| {
                Attempt::Failed(SourceError::Unavailable {
                    message: format!("GitHub GraphQL request failed: {}", transport_error(&e)),
                })
            })?;
        let status = response.status();
        if !status.is_success() {
            *response_outcome = accounting::ResponseOutcome::HttpError;
        }
        let header = |name: &str| whole_seconds(response.headers().get(name));
        let served = response
            .headers()
            .get(reqwest::header::DATE)
            .and_then(|value| value.to_str().ok())
            .and_then(|value| DateTime::parse_from_rfc2822(value).ok())
            .map(|at| at.with_timezone(&Utc));
        *limits = accounting::RateLimit::read(|name| {
            response
                .headers()
                .get(name)
                .and_then(|value| value.to_str().ok())
                .map(str::to_owned)
        });
        // Exactly `0` is exhaustion and everything else — a count, an empty value, bytes
        // that are not text at all — is "not known to be exhausted". This never makes a
        // response a refusal on its own: it says which limiter a refusal is attributed to
        // and where its hint comes from, so a value this cannot read costs a hint rather
        // than an answer.
        let exhausted = response
            .headers()
            .get("x-ratelimit-remaining")
            .and_then(|value| value.to_str().ok())
            == Some("0");
        // `retry-after` is what GitHub asks for when it asks; when it does not and the
        // primary budget is spent, `x-ratelimit-reset` says when that budget comes back,
        // which is the same question answered as an absolute time. Nothing else here is a
        // hint, and a schedule is what answers a refusal that carries none.
        let retry_after = header("retry-after");
        let hinted_by = if retry_after.is_some() || !exhausted {
            Hint::RetryAfter
        } else {
            Hint::PrimaryReset
        };
        let hint = retry_after.or_else(|| {
            exhausted
                .then(|| header("x-ratelimit-reset"))
                .flatten()
                .map(|reset| reset.saturating_sub(Utc::now().timestamp().max(0).unsigned_abs()))
        });
        // Read before it is parsed, because the evidence which tells a secondary rate
        // limit from a rejected credential is in the body of a response whose status says
        // only "forbidden" — and a non-success response was never parsed at all.
        let body = response.text().await.map_err(|e| {
            Attempt::Failed(SourceError::Unavailable {
                message: format!(
                    "GitHub GraphQL response could not be read: {}",
                    transport_error(&e)
                ),
            })
        })?;
        *reported_cost = serde_json::from_str::<Value>(&body)
            .ok()
            .as_ref()
            .and_then(|body| body.pointer("/data/rateLimit/cost"))
            .and_then(Value::as_u64);
        let parsed = serde_json::from_str::<Value>(&body).ok();
        *response_outcome = accounting::ResponseOutcome::of_graphql_reply(status, parsed.as_ref());
        if let Some(limiter) = Limiter::classify(status, exhausted, &body) {
            return Err(Attempt::Limited(Limited {
                limiter,
                hint,
                hinted_by,
            }));
        }
        if status == StatusCode::UNAUTHORIZED || status == StatusCode::FORBIDDEN {
            return Err(Attempt::Failed(SourceError::Auth {
                message: format!(
                    "GitHub rejected the configured credential with HTTP {status}; grant it Projects and Issues read/write plus Pull requests read-only access for every repository represented on the board"
                ),
            }));
        }
        if !status.is_success() {
            return Err(Attempt::Failed(SourceError::Unavailable {
                message: format!("GitHub GraphQL returned HTTP {status}"),
            }));
        }
        let said = refusal_wording(status, &body);
        let normalized = said.to_ascii_lowercase();
        if CONFLICT_WORDINGS
            .iter()
            .any(|wording| normalized.contains(wording))
        {
            return Err(Attempt::Conflicted(said));
        }
        // A lookup is read as the inventoried document it was sized from.
        let query = graphql::inventoried(query);
        let answered = if query == graphql::ISSUES_BY_REFERENCE {
            self.answer_unresolved(&body)
        } else if [
            graphql::CREATE_ISSUES,
            graphql::SEARCH_BATCH,
            graphql::USER_IDS,
            graphql::ADD_COMMENTS,
            graphql::WRITE_TASKS,
            graphql::SET_PARENTS,
            graphql::ADD_RELATES_TO,
            graphql::REMOVE_RELATES_TO,
        ]
        .contains(&query)
        {
            answer_with_errors(&body).map_or_else(|| self.answer(&body), Ok)
        } else {
            self.answer(&body)
        };
        let mut data = answered.map_err(Attempt::Failed)?;
        if let Some(served) = served {
            *self
                .last_served
                .lock()
                .unwrap_or_else(std::sync::PoisonError::into_inner) = Some(served);
            stamp_served(&mut data, &served.to_rfc3339());
        }
        Ok(data)
    }

    /// What one successful HTTP response says, once its GraphQL errors are read.
    /// [`Self::answer`], for a read of issues by reference: a reference GitHub cannot resolve
    /// for this credential — a repository it cannot see, an issue number it does not hold —
    /// is answered `null` beside a `NOT_FOUND` error naming it, and that is an answer about
    /// that reference rather than a failure of the others. Any other error refuses the read,
    /// as there.
    fn answer_unresolved(&self, body: &str) -> Result<Value, SourceError> {
        let parsed: Option<Value> = serde_json::from_str(body).ok();
        // Each error is a `NOT_FOUND` whose path names one slot of the batch, `r<n>` or its
        // `issue`, and the answer holds `null` at exactly that path: anything else — another
        // kind of error, one naming no slot, or one beside a value it says is not there — is a
        // refusal of the whole read.
        let unresolved = |data: &Value, error: &Value| {
            error.get("type").and_then(Value::as_str) == Some("NOT_FOUND")
                && error
                    .get("path")
                    .and_then(Value::as_array)
                    .is_some_and(|path| {
                        let alias = path.first().and_then(Value::as_str).and_then(|alias| {
                            alias
                                .strip_prefix('r')
                                .and_then(|slot| slot.parse::<usize>().ok())
                                .filter(|slot| *slot < REFERENCE_BATCH)
                                .map(|_| alias)
                        });
                        match (alias, path.get(1).and_then(Value::as_str), path.len()) {
                            (Some(alias), None, 1) => data.get(alias).is_some_and(Value::is_null),
                            (Some(alias), Some("issue"), 2) => data
                                .get(alias)
                                .and_then(|repository| repository.get("issue"))
                                .is_some_and(Value::is_null),
                            _ => false,
                        }
                    })
        };
        let unresolved_only = parsed.as_ref().is_some_and(|parsed| {
            parsed.get("data").is_some_and(|data| {
                data.is_object()
                    && parsed
                        .get("errors")
                        .and_then(Value::as_array)
                        .is_some_and(|errors| errors.iter().all(|error| unresolved(data, error)))
            })
        });
        match parsed {
            Some(parsed) if unresolved_only => Ok(parsed["data"].clone()),
            _ => self.answer(body),
        }
    }

    fn answer(&self, body: &str) -> Result<Value, SourceError> {
        let body: Value = serde_json::from_str(body).map_err(|e| SourceError::Malformed {
            message: format!("GitHub returned invalid JSON: {e}"),
        })?;
        let errors = body
            .get("errors")
            .map(|value| {
                value.as_array().ok_or_else(|| SourceError::Malformed {
                    message: "GitHub response errors is not an array".into(),
                })
            })
            .transpose()?;
        if let Some(errors) = errors.filter(|errors| !errors.is_empty()) {
            let messages = errors
                .iter()
                .filter_map(|e| e.get("message").and_then(Value::as_str))
                .collect::<Vec<_>>()
                .join("; ");
            let message = if messages.is_empty() {
                "GitHub returned GraphQL errors".into()
            } else {
                messages
            };
            let normalized = message.to_ascii_lowercase();
            if normalized.contains("resource not accessible") || normalized.contains("scope") {
                return Err(SourceError::Auth {
                    message: format!(
                        "{message}; grant {} Projects and Issues read/write plus Pull requests read-only access for every repository represented on the board",
                        self.credential_name
                    ),
                });
            }
            return Err(SourceError::Refused { message });
        }
        body.get("data")
            .filter(|data| data.is_object())
            .cloned()
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub response has no data object".into(),
            })
    }

    // llmlint: ignore[boundary_inputs_validated] GitHub caps nested connections at 100 and
    // GraphQL cannot independently page them inside the outer item page. This source page is
    // deliberately bounded at that published maximum; the live drift journey exercises it.
    async fn board_page(
        &self,
        items_after: Option<&str>,
        items_first: u32,
    ) -> Result<Value, SourceError> {
        let data = self
            .graphql(
                graphql::BOARD,
                json!({"owner":self.owner,"number":self.project_number,
                       "first":items_first.min(MAX_PAGE_SIZE),"after":items_after,
                       "nestedFirst":NESTED_PAGE_SIZE,"duplicates":true}),
            )
            .await?;
        data.pointer("/owner/projectV2")
            .filter(|v| !v.is_null())
            .cloned()
            .ok_or_else(|| SourceError::Refused {
                message: format!(
                    "GitHub project {}/{} was not found or is not visible to the token",
                    self.owner, self.project_number
                ),
            })
    }

    /// The search that finds the issues of this board, narrowed by `also` when it is
    /// given.
    ///
    /// `project:owner/number` is what scopes a search to one board, and `is:issue` is what
    /// keeps pull requests out of it: GitHub's `ISSUE` search type covers both, and a pull
    /// request is somebody's change rather than a unit of plan. `-has:parent` is *not*
    /// here on purpose — GitHub accepts it and silently ignores it, so a project is told
    /// from a task by the `parent` field each issue carries rather than by the search.
    fn board_search(&self, also: Option<&str>) -> String {
        let scope = format!("project:{}/{} is:issue", self.owner, self.project_number);
        match also {
            Some(also) => format!("{scope} {also}"),
            None => scope,
        }
    }

    /// One issue this source reached directly, as the board item a read of the board would
    /// have produced — or `None` when this board does not hold it.
    ///
    /// The board half of an issue rides along on `Issue.projectItems`, so the value handed
    /// to [`Self::resolve`] is the very shape a `ProjectV2.items` read gives it: the board
    /// item's own id, that item's field values, and the issue as its content. One resolver
    /// for both routes is what makes an issue read through a search, through its own node
    /// id, or through its project's sub-issues report the same title, the same status, the
    /// same labels and the same qualified id.
    ///
    /// An issue with no entry for *this* board is not this source's to report, which is
    /// what keeps an id naming some other repository's issue from being answered as an item
    /// of this board. That answer is given about an **exhausted** connection and never
    /// about an unread page: the entry is looked for on the page in hand, and only if that
    /// page reports more of the connection, in [`Self::board_membership`]'s walk of the
    /// rest of it.
    async fn resolve_issue(&self, issue: &Value) -> Result<Option<Resolved>, SourceError> {
        self.resolve_issue_as(issue, true).await
    }

    /// [`Self::resolve_issue`], remembering what it resolves for this command's later reads
    /// and writes only when `remember` says to.
    async fn resolve_issue_as(
        &self,
        issue: &Value,
        remember: bool,
    ) -> Result<Option<Resolved>, SourceError> {
        if optional_str(issue, "__typename")? != Some("Issue") {
            return Ok(None);
        }
        let memberships = issue
            .get("projectItems")
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub issue is missing projectItems".into(),
            })?;
        let nodes = memberships
            .get("nodes")
            .and_then(Value::as_array)
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub issue projectItems.nodes is not an array".into(),
            })?;
        let held = match self.board_entry(nodes) {
            Some(held) => held.clone(),
            None => {
                let info = memberships
                    .get("pageInfo")
                    .ok_or_else(|| SourceError::Malformed {
                        message: "GitHub issue projectItems has no pageInfo".into(),
                    })?;
                // The page held no entry for this board. Whether that means the issue is
                // not on it is a question about the rest of the connection, and only a
                // connection with no rest answers it here.
                if !required_bool(info, "hasNextPage")? {
                    return Ok(None);
                }
                let cursor = required_str(info, "endCursor")?;
                validate_cursor_progress(None, cursor)?;
                let issue_id = required_str(issue, "id")?;
                match self.board_membership(issue_id, cursor).await? {
                    Some(held) => held,
                    None => return Ok(None),
                }
            }
        };
        // The membership's own `updatedAt` rides along: it is the board half of the item's change
        // token, which a field change moves where the issue's own time does not.
        let item = json!({
            "id": required_str(&held, "id")?,
            "updatedAt": held.get("updatedAt"),
            "project": held.get("project"),
            "fieldValues": held.get("fieldValues"),
            "content": issue,
        });
        if remember {
            self.resolve(&item)
        } else {
            self.resolve_unremembered(&item)
        }
    }

    /// This board's own entry among one page of an issue's `Issue.projectItems`.
    ///
    /// One spelling of *which membership is this board's*, so the page a read carries and
    /// the pages [`Self::board_membership`] walks are searched by the same rule.
    fn board_entry<'a>(&self, nodes: &'a [Value]) -> Option<&'a Value> {
        nodes.iter().find(|node| {
            node.pointer("/project/number").and_then(Value::as_u64)
                == Some(u64::from(self.project_number))
        })
    }

    /// The rest of one issue's board memberships, from `after`, for this board's entry.
    ///
    /// The recovery read: a page of memberships that holds no entry for this board says
    /// nothing about the memberships past it, so the connection is walked to exhaustion
    /// before an issue is reported as one this board does not hold. `Ok(None)` is that
    /// positive answer — the whole connection was read and no entry named this board —
    /// rather than a failure, and the walk is held to
    /// [`validate_cursor_progress`] like every other page walk here, so a source answering
    /// with a cursor that does not advance is refused instead of spun on.
    async fn board_membership(
        &self,
        issue: &str,
        after: &str,
    ) -> Result<Option<Value>, SourceError> {
        let mut after = after.to_owned();
        loop {
            let data = self
                .graphql(
                    graphql::ISSUE_BOARD_ITEMS,
                    json!({"id":issue,"first":MAX_PAGE_SIZE,"after":after,
                           "nestedFirst":NESTED_PAGE_SIZE}),
                )
                .await?;
            let Some(connection) = data
                .pointer("/node/projectItems")
                .filter(|value| !value.is_null())
            else {
                // The id resolved to nothing, or to something with no memberships to walk —
                // which is the same answer as a connection holding no entry for this board.
                return Ok(None);
            };
            let nodes = connection
                .get("nodes")
                .and_then(Value::as_array)
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub issue projectItems.nodes is not an array".into(),
                })?;
            if let Some(held) = self.board_entry(nodes) {
                return Ok(Some(held.clone()));
            }
            let info = connection
                .get("pageInfo")
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub issue projectItems has no pageInfo".into(),
                })?;
            let next = required_bool(info, "hasNextPage")?
                .then(|| required_str(info, "endCursor"))
                .transpose()?;
            match next {
                Some(next) => {
                    validate_cursor_progress(Some(&after), next)?;
                    after = next.to_owned();
                }
                None => return Ok(None),
            }
        }
    }

    /// One page of a board-scoped issue search, and where the next page resumes.
    async fn search_page(
        &self,
        search: &str,
        first: u32,
        after: Option<&str>,
    ) -> Result<(Vec<Resolved>, Option<String>), SourceError> {
        let data = self
            .graphql(
                graphql::SEARCH_ISSUES,
                json!({"search":search,"type":"ISSUE","first":first.min(MAX_PAGE_SIZE),
                       "after":after,"nestedFirst":NESTED_PAGE_SIZE,
                       "boardItems":BOARD_ITEMS_PAGE_SIZE,"duplicates":true}),
            )
            .await?;
        let connection = data.get("search").ok_or_else(|| SourceError::Malformed {
            message: "GitHub search response has no search connection".into(),
        })?;
        let mut found = Vec::new();
        for node in connection
            .get("nodes")
            .and_then(Value::as_array)
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub search nodes is not an array".into(),
            })?
        {
            if let Some(resolved) = self.resolve_issue(node).await? {
                found.push(resolved);
            }
        }
        let info = connection
            .get("pageInfo")
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub search connection has no pageInfo".into(),
            })?;
        let next = required_bool(info, "hasNextPage")?
            .then(|| required_str(info, "endCursor"))
            .transpose()?
            .map(str::to_owned);
        if let Some(next) = &next {
            validate_cursor_progress(after, next)?;
        }
        Ok((found, next))
    }

    /// Every issue this board holds, completed with what this run wrote.
    ///
    /// The completion is not an optimisation and it is not a cache: GitHub's issue search
    /// is an index and is eventually consistent, so an issue this run created seconds ago
    /// can be absent from it, and a project listed straight after being written would
    /// otherwise be missing from its own board. What is added back is only what this
    /// process itself wrote, out of [`Self::created`], which lives and dies with the
    /// process.
    async fn board_issues(&self) -> Result<Vec<Resolved>, SourceError> {
        let found = self.searched_issues().await?;
        self.completed_with_written(found, |_| true)
    }

    /// Every issue this board's own search reports, walked to exhaustion, read once per
    /// source.
    ///
    /// The uncompleted half of [`Self::board_issues`], separated because [`Self::board`]
    /// needs it too and the two would otherwise walk the same search twice in one command.
    /// See [`Self::search_cache`] for why holding it is the same bargain holding the board
    /// is.
    async fn searched_issues(&self) -> Result<Vec<Resolved>, SourceError> {
        let cached = self.search_cache()?.clone();
        if let Some(held) = cached {
            return Ok(held);
        }
        let mut after: Option<String> = None;
        let mut found = Vec::new();
        let search = self.board_search(None);
        loop {
            let (page, next) = self
                .search_page(&search, MAX_PAGE_SIZE, after.as_deref())
                .await?;
            found.extend(page);
            match next {
                Some(next) => after = Some(next),
                None => break,
            }
        }
        *self.search_cache()? = Some(found.clone());
        Ok(found)
    }

    /// Every issue this board's own search reports with `also` beside its scope, walked to
    /// exhaustion — held for no later read, which may not be narrowed alike.
    async fn searched_narrowly(&self, also: &str) -> Result<Vec<Resolved>, SourceError> {
        let mut after: Option<String> = None;
        let mut found = Vec::new();
        let search = self.board_search(Some(also));
        loop {
            let (page, next) = self
                .search_page(&search, MAX_PAGE_SIZE, after.as_deref())
                .await?;
            found.extend(page);
            match next {
                Some(next) => after = Some(next),
                None => return Ok(found),
            }
        }
    }

    /// This process's own view of the board's issues, or the refusal a poisoned lock is.
    fn search_cache(
        &self,
    ) -> Result<std::sync::MutexGuard<'_, Option<Vec<Resolved>>>, SourceError> {
        self.search_cache
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "this source's view of the board's issues was left inconsistent by an \
                      earlier failure; next: run the command again"
                    .into(),
            })
    }

    /// `found`, with everything this run wrote that `keep` accepts and the read did not
    /// report.
    ///
    /// See [`Self::created`] and [`Self::board_issues`] for why a read has to be completed
    /// at all: the search index is behind, and a node read of an item filed moments ago can
    /// be too.
    fn completed_with_written(
        &self,
        mut found: Vec<Resolved>,
        keep: impl Fn(&Resolved) -> bool,
    ) -> Result<Vec<Resolved>, SourceError> {
        for own in self.created()?.iter().filter(|own| keep(own)) {
            if !found.iter().any(|item| item.id == own.id) {
                found.push(own.clone());
            }
        }
        Ok(found)
    }

    /// What resolving one node id reached.
    ///
    /// Three answers rather than an `Option`, because a board *draft* is none of the other
    /// two: it is not an issue, so the issue fragment reads nothing of it, and a read of one
    /// is completed by a read of the draft itself rather than reported as nothing.
    async fn reach(&self, id: &NativeId) -> Result<Reached, SourceError> {
        let asked = self
            .graphql(
                graphql::ISSUE,
                json!({"id":id.0,"first":MAX_PAGE_SIZE,"nestedFirst":NESTED_PAGE_SIZE,
                       "boardItems":BOARD_ITEMS_PAGE_SIZE,"duplicates":true}),
            )
            .await;
        let data = match asked {
            Ok(data) => data,
            // A string that is not a node id at all is not a failure to report: it is an id
            // this board does not hold, which is what every read of one already answers.
            Err(error) if unresolvable_node(&error) => return Ok(Reached::Nothing),
            Err(error) => return Err(error),
        };
        let Some(node) = data.get("node").filter(|value| !value.is_null()) else {
            return Ok(Reached::Nothing);
        };
        if optional_str(node, "__typename")? == Some("DraftIssue") {
            return Ok(Reached::Draft);
        }
        Ok(match self.resolve_issue(node).await? {
            Some(item) => Reached::Held(Box::new(item)),
            None => Reached::Nothing,
        })
    }

    /// One item of this board by its own id, whatever kind it is.
    ///
    /// Resolved from the identifier alone: no search, board-wide or otherwise. What this
    /// run wrote is read first, because a node read of an item created moments ago can
    /// still be behind the board field values written onto it — see [`Self::created`].
    async fn item_by_id(&self, id: &NativeId) -> Result<Option<Resolved>, SourceError> {
        if let Some(own) = self.created()?.iter().find(|own| own.id == *id) {
            return Ok(Some(own.clone()));
        }
        if let Some(held) = self.read_ahead()?.remove(id) {
            return Ok(held);
        }
        match self.reach(id).await? {
            Reached::Held(item) => Ok(Some(*item)),
            Reached::Nothing => Ok(None),
            Reached::Draft => self.draft_by_id(id).await,
        }
    }

    /// [`Self::item_by_id`] for a document: one [`graphql::DOCUMENT_IN_PROJECT`] read, which
    /// answers the project it is filed under and that project's members beside it.
    ///
    /// Each is held for this command as a read of it would have been: the project for the
    /// next read of it by its id, and — when its page of members is the whole of them — the
    /// members for the next listing of that project. So a write held to who may read the
    /// document, which reads all three, is answered by this one request.
    async fn document_by_id(&self, id: &NativeId) -> Result<Option<Resolved>, SourceError> {
        if self.created()?.iter().any(|own| own.id == *id) || self.read_ahead()?.contains_key(id) {
            return self.item_by_id(id).await;
        }
        let asked = self
            .graphql(
                graphql::DOCUMENT_IN_PROJECT,
                json!({"id":id.0,"first":MAX_PAGE_SIZE,"memberFirst":MEMBER_PAGE_SIZE,
                       "nestedFirst":NESTED_PAGE_SIZE,"boardItems":BOARD_ITEMS_PAGE_SIZE,
                       "duplicates":true}),
            )
            .await;
        let data = match asked {
            Ok(data) => data,
            Err(error) if unresolvable_node(&error) => return Ok(None),
            Err(error) => return Err(error),
        };
        let Some(node) = data.get("node").filter(|value| !value.is_null()) else {
            return Ok(None);
        };
        if optional_str(node, "__typename")? == Some("DraftIssue") {
            return self.draft_by_id(id).await;
        }
        let Some(item) = self.resolve_issue(node).await? else {
            return Ok(None);
        };
        // An answer for another issue than the one asked about says nothing of that one.
        if item.id != *id {
            return Err(SourceError::Malformed {
                message: format!(
                    "GitHub answered issue {} when asked for document {}",
                    item.id.0, id.0
                ),
            });
        }
        if let Some(project) = node.get("project").filter(|value| !value.is_null())
            && let Some(parent) = self
                .resolve_issue(project)
                .await?
                .filter(|parent| Some(&parent.id) == item.parent.as_ref())
        {
            let members = project.get("subIssues").filter(|value| !value.is_null());
            let whole = members
                .and_then(|members| members.pointer("/pageInfo/hasNextPage"))
                .and_then(Value::as_bool)
                == Some(false);
            if whole {
                let nodes = members
                    .and_then(|members| members.get("nodes"))
                    .and_then(Value::as_array)
                    .ok_or_else(|| SourceError::Malformed {
                        message: "GitHub subIssues.nodes is not an array".into(),
                    })?;
                let mut children = Vec::new();
                for member in nodes {
                    if let Some(resolved) = self.resolve_issue(member).await? {
                        children.push(resolved);
                    }
                }
                // Held as the project's members only when every one resolved and says it is
                // filed under it: a page with any other answers nothing of all the members.
                if children.len() == nodes.len()
                    && children
                        .iter()
                        .all(|child| child.parent.as_ref() == Some(&parent.id))
                {
                    self.children_cache()?
                        .insert(parent.id.clone(), (parent.id.clone(), children));
                }
            }
            self.read_ahead()?.insert(parent.id.clone(), Some(parent));
        }
        Ok(Some(item))
    }

    /// Several items of this board, each by its own id, in order — what [`Self::item_by_id`]
    /// answers for each, read [`DETAIL_BATCH`] at a time with [`graphql::ISSUE_DETAILS`] rather
    /// than one request per id.
    ///
    /// What this run wrote answers first, as it does there, and only the rest is read. One id
    /// left to read is read by [`Self::item_by_id`] itself, which costs what a batch does. A
    /// batch GitHub refuses because one of its ids resolves to no node at all is read again one
    /// id at a time, so that id is answered as not held and the others as themselves; a draft
    /// is completed by a read of the draft, exactly as there.
    async fn items_by_ids(&self, ids: &[NativeId]) -> Result<Vec<Option<Resolved>>, SourceError> {
        let mut found: Vec<Option<Option<Resolved>>> = {
            let created = self.created()?;
            ids.iter()
                .map(|id| {
                    created
                        .iter()
                        .find(|own| own.id == *id)
                        .map(|own| Some(own.clone()))
                })
                .collect()
        };
        let unread: Vec<NativeId> = ids
            .iter()
            .zip(&found)
            .filter(|(_, found)| found.is_none())
            .map(|(id, _)| id.clone())
            .collect();
        let mut read = Vec::with_capacity(unread.len());
        if let [one] = unread.as_slice() {
            read.push(self.item_by_id(one).await?);
        } else {
            for batch in unread.chunks(DETAIL_BATCH) {
                let data = match self
                    .graphql(graphql::ISSUE_DETAILS, detail_batch(batch, None))
                    .await
                {
                    Ok(data) => data,
                    Err(error) if unresolvable_node(&error) => {
                        for id in batch {
                            read.push(self.item_by_id(id).await?);
                        }
                        continue;
                    }
                    Err(error) => return Err(error),
                };
                for (slot, id) in batch.iter().enumerate() {
                    let node =
                        data.get(format!("i{slot}"))
                            .ok_or_else(|| SourceError::Malformed {
                                message: format!(
                                    "GitHub answered a batch read with no item for {}",
                                    id.0
                                ),
                            })?;
                    read.push(if node.is_null() {
                        None
                    } else if optional_str(node, "__typename")? == Some("DraftIssue") {
                        self.draft_by_id(id).await?
                    } else {
                        if optional_str(node, "__typename")? == Some("Issue")
                            && required_str(node, "id")? != id.0
                        {
                            return Err(SourceError::Malformed {
                                message: format!(
                                    "GitHub answered the read of {} with issue {}",
                                    id.0,
                                    required_str(node, "id")?
                                ),
                            });
                        }
                        self.resolve_issue(node).await?
                    });
                }
            }
        }
        let mut read = read.into_iter();
        Ok(found
            .iter_mut()
            .map(|slot| slot.take().unwrap_or_else(|| read.next().flatten()))
            .collect())
    }

    fn read_ahead(
        &self,
    ) -> Result<std::sync::MutexGuard<'_, BTreeMap<NativeId, Option<Resolved>>>, SourceError> {
        self.read_ahead
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message:
                    "items read ahead for a copy were left inconsistent; run the command again"
                        .into(),
            })
    }

    fn resolved_cache(
        &self,
    ) -> Result<std::sync::MutexGuard<'_, BTreeMap<NativeId, Resolved>>, SourceError> {
        self.resolved_cache
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "resolved item records were left inconsistent; run the command again"
                    .into(),
            })
    }

    /// Reuse a record this invocation already resolved. The mutation sender invalidates
    /// it before writing, so a partial failure cannot leave a pre-write binding behind.
    async fn bound_item(&self, id: &NativeId) -> Result<Option<Resolved>, SourceError> {
        let cached = self.resolved_cache()?.get(id).cloned();
        if let Some(item) = cached {
            return Ok(Some(item));
        }
        self.item_by_id(id).await
    }

    /// Take out the item an issue read by reference reached for `id`, when one is held: the
    /// status write it was read for is the one caller, so no other read or write is ever
    /// answered from a read that selected only what that write needs.
    fn take_held(&self, id: &NativeId) -> Result<Option<Resolved>, SourceError> {
        Ok(self
            .held_cache
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "items read by issue reference were left inconsistent; run the \
                          command again"
                    .into(),
            })?
            .remove(id))
    }

    /// One board draft by its own id, with the board item it sits in — or `None` when no
    /// item of this board is that draft's.
    ///
    /// The same decision [`Self::resolve_issue`] makes for an issue, over the draft's own
    /// `projectV2Items`: an entry naming this board is what makes it this board's. GitHub
    /// links a draft to one board item, so the page this read carries is the whole of that
    /// connection, and a page that reports more than it holds is refused rather than read
    /// as an answer about memberships nobody read.
    async fn draft_by_id(&self, id: &NativeId) -> Result<Option<Resolved>, SourceError> {
        let data = self
            .graphql(
                graphql::DRAFT,
                json!({"id":id.0,"nestedFirst":NESTED_PAGE_SIZE,
                       "boardItems":BOARD_ITEMS_PAGE_SIZE}),
            )
            .await?;
        // Gone between the two reads is an answer — the draft is no longer there. Anything
        // else than the draft [`Self::reach`] was just told this id is, is not one.
        let Some(draft) = data.get("node").filter(|node| !node.is_null()) else {
            return Ok(None);
        };
        if optional_str(draft, "__typename")? != Some("DraftIssue") {
            return Err(SourceError::Malformed {
                message: format!(
                    "GitHub answered {} as a draft and then as something else",
                    id.0
                ),
            });
        }
        if required_str(draft, "id")? != id.0 {
            return Err(SourceError::Malformed {
                message: format!("GitHub answered a different draft for {}", id.0),
            });
        }
        let memberships = draft
            .get("projectV2Items")
            .ok_or_else(|| SourceError::Malformed {
                message: format!("GitHub draft {} is missing projectV2Items", id.0),
            })?;
        let nodes = memberships
            .get("nodes")
            .and_then(Value::as_array)
            .ok_or_else(|| SourceError::Malformed {
                message: format!("GitHub draft {} projectV2Items.nodes is not an array", id.0),
            })?;
        let info = memberships
            .get("pageInfo")
            .ok_or_else(|| SourceError::Malformed {
                message: format!("GitHub draft {} projectV2Items has no pageInfo", id.0),
            })?;
        // Read whether or not this board's entry is on the page: a page claiming more than
        // the one item GitHub links a draft to is a malformed answer either way.
        if required_bool(info, "hasNextPage")? || nodes.len() > 1 {
            return Err(SourceError::Malformed {
                message: format!(
                    "GitHub draft {} reports more board items than the one GitHub links a draft \
                     to",
                    id.0
                ),
            });
        }
        if let Some(node) = nodes.first()
            && node
                .pointer("/project/number")
                .and_then(Value::as_u64)
                .is_none()
        {
            return Err(SourceError::Malformed {
                message: format!(
                    "GitHub draft {} board item has no numeric project number",
                    id.0
                ),
            });
        }
        let Some(held) = self.board_entry(nodes) else {
            return Ok(None);
        };
        if required_str(
            held.get("project").ok_or_else(|| SourceError::Malformed {
                message: format!("GitHub draft {} board item has no project", id.0),
            })?,
            "id",
        )? != self.board_fields().await?.id.as_str()
        {
            return Ok(None);
        }
        let item = json!({
            "id": required_str(held, "id")?,
            "project": held.get("project"),
            "fieldValues": held.get("fieldValues"),
            "content": draft,
        });
        self.resolve(&item)
    }

    /// The board's own id and field definitions, for a write whose item does not carry
    /// them — never its items.
    ///
    /// A board this command has already listed supplies them, since it read them beside its
    /// items; otherwise they come from [`graphql::BOARD_FIELDS`], once per command. Neither
    /// is consulted about which items the board holds: see the module documentation for
    /// why a question about one known item is answered by reading that item.
    async fn board_fields(&self) -> Result<BoardFields, SourceError> {
        if let Some(board) = self.board_cache()?.as_ref() {
            return Ok(BoardFields {
                id: BoardId::parse(&board.id)?,
                fields: board.fields.clone(),
            });
        }
        if let Some(held) = self.fields_cache()?.clone() {
            return Ok(held);
        }
        let data = self
            .graphql(
                graphql::BOARD_FIELDS,
                json!({"owner":self.owner,"number":self.project_number,
                       "nestedFirst":NESTED_PAGE_SIZE}),
            )
            .await?;
        self.fields_read(&data)
    }

    /// The board's id and fields out of an answer carrying the `boardFields` root, held for
    /// the rest of this command.
    fn fields_read(&self, data: &Value) -> Result<BoardFields, SourceError> {
        let board = data
            .pointer("/boardFields/projectV2")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Refused {
                message: format!(
                    "GitHub project {}/{} was not found or is not visible to the token",
                    self.owner, self.project_number
                ),
            })?;
        let read = BoardFields {
            id: BoardId::parse(required_str(board, "id")?)?,
            fields: board.get("fields").cloned().unwrap_or(Value::Null),
        };
        *self.fields_cache()? = Some(read.clone());
        Ok(read)
    }

    /// Read what creating an issue in `repository` needs and this command has not read yet —
    /// the board's fields and the repository's node id — in one request when it needs both.
    ///
    /// When either is already known this sends nothing, and the other is read by its own
    /// document where it is asked for, so no create reads anything twice.
    async fn creation_context(
        &self,
        repository: &RepositoryTarget,
        incoming: &Incoming<'_>,
    ) -> Result<(), SourceError> {
        let fields_known = self.board_cache()?.is_some() || self.fields_cache()?.is_some();
        if fields_known || self.repository_cache()?.contains_key(repository) {
            return Ok(());
        }
        // A placement an earlier process kept is taken only while the project issue this create
        // has just read shows its setup-context token unchanged; see `projection`.
        if self.placed_by_kept(repository, incoming.parent) {
            return Ok(());
        }
        let data = self
            .graphql(
                graphql::CREATION_CONTEXT,
                json!({"owner":self.owner,"number":self.project_number,
                       "nestedFirst":NESTED_PAGE_SIZE,"repositoryOwner":repository.owner,
                       "repositoryName":repository.name}),
            )
            .await?;
        self.fields_read(&data)?;
        self.repository_read(&data, repository, incoming)?;
        Ok(())
    }

    /// This process's own view of the board's fields, or the refusal a poisoned lock is.
    fn fields_cache(&self) -> Result<std::sync::MutexGuard<'_, Option<BoardFields>>, SourceError> {
        self.fields_cache
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "this source's view of the board's fields was left inconsistent by an \
                      earlier failure; next: run the command again"
                    .into(),
            })
    }

    /// What a write to `item` needs of the board, read off that item when it says enough and
    /// off [`Self::board_fields`] when it does not.
    ///
    /// A node read of an item names its board and carries the definition of every field it
    /// holds a value of — so an item naming its board, holding a value of the origin field
    /// when the write carries an origin, and, when the write carries a status, holding a
    /// `Status` value, needs no read of the board at all. **Nothing the item does not say is
    /// guessed:** a field it holds no value of may still be on the board, and a view reading it
    /// as absent would refuse a write the board can take or skip a field write the board
    /// needs, so such an item — and a create, which has no item yet — takes the board's fields
    /// from their own read instead. A write carrying no origin to an item holding none — every
    /// write out of a source declared private — writes no origin whatever the board defines,
    /// so it needs no definition of that field.
    async fn fields_for(
        &self,
        item: Option<&Resolved>,
        writes_origin: bool,
        writes_status: bool,
        selects_priority: bool,
    ) -> Result<BoardFields, SourceError> {
        if let Some(board) = item.and_then(Resolved::carried_board) {
            return Ok(board);
        }
        if let Some(item) = item
            && let Some(board_id) = item.named_board()
            && (!writes_origin || item.defines(ORIGIN_FIELD))
            && (!writes_status || item.defines("Status"))
            && (!selects_priority || item.defines(PRIORITY_FIELD))
            && self
                .metadata_fields
                .iter()
                .all(|projection| item.defines(&projection.field))
        {
            return Ok(BoardFields {
                id: board_id,
                fields: json!({"nodes": item.fields, "pageInfo": {"hasNextPage": false}}),
            });
        }
        self.board_fields().await
    }

    /// Everything filed under one issue of this board, walked to exhaustion — or `None`
    /// when that id names nothing here with a sub-issue relationship to walk.
    ///
    /// `None` and an empty answer are different: `None` is *this is not an issue of this
    /// GitHub*, which is what sends a project selector on to be read as a name, and an
    /// empty vector is a project that holds nothing.
    async fn sub_issues(&self, id: &NativeId) -> Result<Option<Vec<Resolved>>, SourceError> {
        let mut after: Option<String> = None;
        let mut children = Vec::new();
        // The project issue as the first page carried it, held once the walk has ended.
        let mut parent: Option<Resolved> = None;
        loop {
            let asked = self
                .graphql(
                    graphql::SUB_ISSUES,
                    json!({"id":id.0,"memberFirst":MEMBER_PAGE_SIZE,"after":after,
                           "nestedFirst":NESTED_PAGE_SIZE,
                           "boardItems":BOARD_ITEMS_PAGE_SIZE,"duplicates":true}),
                )
                .await;
            let data = match asked {
                Ok(data) => data,
                // A string that is not a node id at all is not a failure to report: it is
                // the ordinary answer to a selector naming a project by its name.
                Err(error) if unresolvable_node(&error) => return Ok(None),
                Err(error) => return Err(error),
            };
            let Some(connection) = data
                .pointer("/node/subIssues")
                .filter(|value| !value.is_null())
            else {
                // No such node, or one with no sub-issue relationship — a board draft is
                // the one this board can really hold.
                return Ok(None);
            };
            if after.is_none()
                && let Some(node) = data.get("node").filter(|node| node.get("id").is_some())
            {
                parent = self
                    .resolve_issue(node)
                    .await?
                    .filter(|resolved| resolved.id == *id);
            }
            for node in connection
                .get("nodes")
                .and_then(Value::as_array)
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub subIssues.nodes is not an array".into(),
                })?
            {
                if let Some(resolved) = self.resolve_issue(node).await? {
                    children.push(resolved);
                }
            }
            let info = connection
                .get("pageInfo")
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub subIssues connection has no pageInfo".into(),
                })?;
            let next = required_bool(info, "hasNextPage")?
                .then(|| required_str(info, "endCursor"))
                .transpose()?;
            match next {
                Some(next) => {
                    validate_cursor_progress(after.as_deref(), next)?;
                    after = Some(next.to_owned());
                }
                None => {
                    // Held only once the walk has ended, so one failing part way holds
                    // nothing and every read it would have answered is made.
                    // The project issue the first page carried answers the next read of the
                    // project by its id in this command, as a read of it made now would: what
                    // classifying a project for a write reads straight after its members.
                    if let Some(parent) = &parent {
                        self.read_ahead()?
                            .entry(parent.id.clone())
                            .or_insert_with(|| Some(parent.clone()));
                    }
                    let mut carried = self.carried()?;
                    for item in parent.into_iter().chain(children.iter().cloned()) {
                        carried.insert(item.id.clone(), item);
                    }
                    return Ok(Some(children));
                }
            }
        }
    }

    fn carried(
        &self,
    ) -> Result<std::sync::MutexGuard<'_, BTreeMap<NativeId, Resolved>>, SourceError> {
        self.carried.lock().map_err(|_| SourceError::Unavailable {
            message: "items a project's listing carried were left inconsistent; run the \
                      command again"
                .into(),
        })
    }

    /// Which issue of this board a project *name* is, or `None` when none is.
    ///
    /// One bounded query which filters on that name at the server, rather than a walk of
    /// every issue the board holds. The name is compared again here: the qualifier narrows
    /// what GitHub sends, and this source decides what it names.
    async fn project_by_name(&self, name: &str) -> Result<Option<NativeId>, SourceError> {
        let search = self.board_search(Some(&title_qualifier(name)));
        let mut after = None;
        loop {
            let (candidates, next) = self
                .search_page(&search, SEARCH_PAGE_SIZE, after.as_deref())
                .await?;
            if let Some(item) = candidates.into_iter().find(|item| {
                item.kind == BoardKind::Work(ItemKind::Project)
                    && item.title.eq_ignore_ascii_case(name)
            }) {
                return Ok(Some(item.id));
            }
            match next {
                Some(next) => after = Some(next),
                None => return Ok(None),
            }
        }
    }

    /// Everything filed under one project of this board: the sub-issues of the issue that
    /// project is.
    ///
    /// Tasks *and* documents, because a document filed under a project is a sub-issue of it
    /// too — the caller keeps the kind it asked for. Nothing about this grows as the board
    /// gains projects, or as another project gains tasks.
    ///
    /// A qualified id names the issue and is asked for its sub-issues directly: one
    /// request, no search of any kind. Only a selector GitHub cannot resolve that way is
    /// read as a project *name*, which costs the one bounded search
    /// [`Self::project_by_name`] makes.
    ///
    /// What GitHub answered is held for the rest of the command — see [`Self::children_cache`]
    /// — and each answer is still cut to the items whose parent is this project, so one this
    /// process has since filed elsewhere is not reported here, and completed with what this
    /// process filed under it.
    async fn project_children(&self, selector: &NativeId) -> Result<Vec<Resolved>, SourceError> {
        let held = self.children_cache()?.get(selector).cloned();
        let (project, children) = match held {
            Some(held) => held,
            None => {
                let answered = match self.sub_issues(selector).await? {
                    Some(children) => (selector.clone(), children),
                    None => match self.project_by_name(&selector.0).await? {
                        Some(project) => {
                            let children = self.sub_issues(&project).await?.unwrap_or_default();
                            (project, children)
                        }
                        None => return Ok(Vec::new()),
                    },
                };
                self.children_cache()?
                    .insert(selector.clone(), answered.clone());
                answered
            }
        };
        let mut children: Vec<Resolved> = children
            .into_iter()
            .filter(|child| child.parent.as_ref() == Some(&project))
            .collect();
        for own in self.updated()?.iter() {
            if own.parent.as_ref() == Some(&project) && !children.iter().any(|c| c.id == own.id) {
                children.push(own.clone());
            }
        }
        self.completed_with_written(children, |own| own.parent.as_ref() == Some(&project))
    }

    /// Every issue of this board GitHub's issue search reports updated at or after `since`,
    /// completed with what this run wrote — the candidates a comment-activity read confirms.
    ///
    /// Scoped by the board and by nothing else: `project:<owner>/<number>` reaches every issue
    /// on the board whatever repository, and whatever owner, it lives in, so no repository or
    /// owner qualifier is added and none is needed. What makes the `updated:` qualifier
    /// sufficient is a fact about GitHub rather than about this source: a comment written on an
    /// issue **and a comment edited on it** both move that issue's `updatedAt`. The credentialed
    /// journey `an_edited_comment_moves_its_issue_and_is_selected_since` in `tests/journey`
    /// re-takes that fact on every run of the lane, so a change on GitHub's side fails there
    /// rather than silently narrowing a caller's answer.
    ///
    /// The instant is written to the second, rounded down, which can only widen what the
    /// search returns; confirmation against each candidate's own comments is what makes the
    /// answer exact. The search is an index that lags a write by a second or two — the module
    /// documentation records it — so a caller that asks again from its last instant should
    /// overlap the two by more than that.
    async fn updated_since(&self, since: DateTime<Utc>) -> Result<Vec<Resolved>, SourceError> {
        let found = self.searched(&updated_qualifier(since)).await?;
        self.completed_with_written(found, |_| true)
    }

    /// Every issue of this board GitHub's issue search reports for the board-scoped search
    /// narrowed by `also`, in pages of [`SEARCH_PAGE_SIZE`].
    ///
    /// Uncompleted: what this process wrote is added by the caller, which knows whether its
    /// own record is the fresher of the two.
    async fn searched(&self, also: &str) -> Result<Vec<Resolved>, SourceError> {
        let search = self.board_search(Some(also));
        let mut after: Option<String> = None;
        let mut found = Vec::new();
        loop {
            let (page, next) = self
                .search_page(&search, SEARCH_PAGE_SIZE, after.as_deref())
                .await?;
            found.extend(page);
            match next {
                Some(next) => after = Some(next),
                None => return Ok(found),
            }
        }
    }

    /// A bounded task answer; the versioned cursor carries the connection position, how
    /// many rows of the page starting there were already handed out, and the own-write ids
    /// already observed, including across a new source instance.
    ///
    /// Every page is sent at [`SEARCH_PAGE_SIZE`] whatever the caller's limit, and a limit is
    /// sliced from the pages it needs; why is the module documentation's paging contract.
    async fn search_tasks(
        &self,
        query: &TaskQuery,
        page: &PageRequest,
        also: &str,
    ) -> Result<Page<Task>, SourceError> {
        let mut position = match &page.cursor {
            None => SearchPosition::default(),
            Some(cursor) => serde_json::from_str::<SearchPosition>(&cursor.0)
                .ok()
                .filter(|position| {
                    position.version == SEARCH_CURSOR_VERSION
                        && position.connection.valid_resume(position.offset)
                })
                .ok_or_else(|| SourceError::Config {
                    message: "page cursor is invalid".into(),
                })?,
        };
        let search = self.board_search(Some(also));
        let limit = page.limit.min(MAX_PAGE_SIZE) as usize;
        let own = self.with_own_writes(Vec::new())?;
        // An issue this process commented on is a candidate of a comment-activity read
        // whether or not the search has caught up with the comment; see `Self::commented`.
        let commented = match query.commented_since {
            Some(_) => self.commented()?.clone(),
            None => Vec::new(),
        };
        for id in own.iter().map(|item| &item.id).chain(&commented) {
            if !position.own.contains(id) {
                position.own.push(id.clone());
            }
        }
        let mut tasks = Vec::new();
        while !position.connection.exhausted() && tasks.len() < limit {
            let first = SEARCH_PAGE_SIZE;
            // Page size is part of the key: a short cached answer cannot answer a wider ask.
            let key =
                serde_json::to_string(&("page", &search, &position.connection.after(), first))
                    .expect("search page key is serializable");
            let cached = if query.commented_since.is_none() {
                self.narrowed_cache()?.get(&key).cloned()
            } else {
                None
            };
            let (found, next) = match cached {
                Some(found) => {
                    let next = self
                        .search_next
                        .lock()
                        .map_err(|_| SourceError::Unavailable {
                            message:
                                "search pagination was left inconsistent; run the command again"
                                    .into(),
                        })?
                        .get(&key)
                        .cloned()
                        .flatten();
                    (found, next)
                }
                None => {
                    let (found, next) = self
                        .search_page(&search, first, position.connection.after())
                        .await?;
                    if query.commented_since.is_none() {
                        self.search_next
                            .lock()
                            .map_err(|_| SourceError::Unavailable {
                                message:
                                    "search pagination was left inconsistent; run the command again"
                                        .into(),
                            })?
                            .insert(key.clone(), next.clone());
                        self.narrowed_cache()?.insert(key, found.clone());
                    }
                    (found, next)
                }
            };
            let rows = found.len();
            // Every candidate of the page this read confirms by its comments is read together.
            let candidates =
                self.comment_candidates(found.iter().skip(position.offset), query, &query.project)?;
            self.hold_comment_pages(&candidates).await?;
            for mut item in found.into_iter().skip(position.offset) {
                if tasks.len() == limit {
                    break;
                }
                position.offset += 1;
                if position.own.contains(&item.id) {
                    if position.seen.contains(&item.id) {
                        continue;
                    }
                    position.seen.push(item.id.clone());
                    // The search's own copy of an issue this process only commented on is as
                    // good as a node read of it, since its comments are read either way.
                    let only_commented = commented.contains(&item.id)
                        && !own.iter().any(|written| written.id == item.id);
                    if !only_commented {
                        let updated_at = item.updated_at;
                        let Some(written) = self.search_written(&own, &item.id).await? else {
                            continue;
                        };
                        item = written;
                        item.updated_at = item.updated_at.max(updated_at);
                        self.resolved_cache()?.insert(item.id.clone(), item.clone());
                    }
                }
                if item.kind == BoardKind::Work(ItemKind::Task) {
                    let task = item.task()?;
                    if task_matches(&task, query, &query.project)
                        && self.commented_since(&item, query.commented_since).await?
                    {
                        tasks.push(task);
                    }
                }
            }
            if position.offset < rows {
                continue;
            }
            position.offset = 0;
            position.connection = match next {
                Some(after) => SearchConnection::Continuing {
                    after: Cursor(after),
                },
                None => SearchConnection::Exhausted {},
            };
        }
        if position.connection.exhausted() {
            for id in position.own.clone() {
                if position.seen.contains(&id) {
                    continue;
                }
                if tasks.len() == limit {
                    break;
                }
                position.seen.push(id.clone());
                let Some(item) = self.search_written(&own, &id).await? else {
                    continue;
                };
                if item.kind == BoardKind::Work(ItemKind::Task) {
                    let task = item.task()?;
                    if task_matches(&task, query, &query.project)
                        && self.commented_since(&item, query.commented_since).await?
                    {
                        tasks.push(task);
                    }
                }
            }
        }
        let more = !position.connection.exhausted()
            || position.own.iter().any(|id| !position.seen.contains(id));
        Ok(Page {
            items: tasks,
            next: more.then(|| {
                Cursor(serde_json::to_string(&position).expect("search position is serializable"))
            }),
        })
    }

    /// A resumed process has the ids but no write records; resolve only a record the
    /// current page needs, by its uncached node read rather than the lagging search index.
    async fn search_written(
        &self,
        own: &[Resolved],
        id: &NativeId,
    ) -> Result<Option<Resolved>, SourceError> {
        match own.iter().find(|item| item.id == *id) {
            Some(item) => Ok(Some(item.clone())),
            None => self.item_by_id(id).await,
        }
    }

    /// The candidates for a task query carrying a text, metadata or origin predicate, read
    /// without enumerating the board — or `None` for a query carrying none of the three, which
    /// keeps the reads it always had.
    ///
    /// An origin is answered by [`Self::origin_candidates`], whatever else the query carries,
    /// because it names at most a handful of items. Text and metadata are answered by one
    /// board-scoped issue search carrying every term — see [`narrowing_qualifiers`] — narrowed
    /// further by `updated:>=` when the query also asks for comment activity, since both
    /// qualifiers must hold of an issue the answer keeps. Every candidate is confirmed in
    /// process afterwards by the same predicates [`task_matches`] applies to every read.
    ///
    /// Completed with what this process wrote, its own record winning over the index's copy
    /// of the same item: see [`Self::with_own_writes`].
    async fn narrowed(&self, query: &TaskQuery) -> Result<Option<Vec<Resolved>>, SourceError> {
        self.send_pending_origins(None).await?;
        let asked = match (&query.origin, narrowing_qualifiers(query)) {
            (Some(origin), _) => Narrowing::Origin(origin.clone()),
            (None, Some(qualifiers)) => Narrowing::Search(match query.commented_since {
                Some(since) => format!("{} {qualifiers}", updated_qualifier(since)),
                None => qualifiers,
            }),
            (None, None) => return Ok(None),
        };
        // A question about comment activity is asked afresh every time, as it always was: it
        // is the one a caller polls from one source while waiting for the index, and an
        // answer held from the first poll would be the answer to every later one.
        let key = query.commented_since.is_none().then(|| asked.key());
        let cached = match &key {
            Some(key) => self.narrowed_cache()?.get(key).cloned(),
            None => None,
        };
        let found = match cached {
            Some(found) => found,
            None => {
                let found = match &asked {
                    Narrowing::Origin(origin) => self.origin_candidates(origin).await?,
                    Narrowing::Search(also) => self.searched(also).await?,
                };
                if let Some(key) = key {
                    self.narrowed_cache()?.insert(key, found.clone());
                }
                found
            }
        };
        self.with_own_writes(found).map(Some)
    }

    /// The candidates for a project or unscoped document query carrying a searchable text,
    /// read without enumerating the board — or `None` for a query with no text or a blank one,
    /// which keeps the read it always had.
    ///
    /// The text is sent as the very phrase a task query's text is — see [`text_qualifiers`] —
    /// in one board-scoped issue search walked to its end at [`SEARCH_PAGE_SIZE`], so what it
    /// costs is the issues that match and never the board. Its answer is held for the command
    /// under the same key [`Self::narrowed`] holds that search under, so a walk of the caller's
    /// pages asks GitHub once. Every candidate is confirmed afterwards by its kind and by the
    /// substring rule, exactly as an item of the wider read was, and is completed with what this
    /// process wrote: see [`Self::with_own_writes`].
    async fn text_searched(
        &self,
        text: Option<&TextQuery>,
    ) -> Result<Option<Vec<Resolved>>, SourceError> {
        let Some(also) = text_qualifiers(text) else {
            return Ok(None);
        };
        let key = Narrowing::Search(also.clone()).key();
        let cached = self.narrowed_cache()?.get(&key).cloned();
        let found = match cached {
            Some(found) => found,
            None => {
                let found = self.searched(&also).await?;
                self.narrowed_cache()?.insert(key, found.clone());
                found
            }
        };
        self.with_own_writes(found).map(Some)
    }

    /// Every `kind` item of this board whose own origin is exactly `origin`: the candidates
    /// [`Self::narrowed`] reads for a task query carrying that origin — one
    /// [`graphql::ORIGIN_LOOKUP`] walk, held for the command under the same key, completed with
    /// what this process wrote — each confirmed against its own origin entry.
    ///
    /// What answers a project or document listing by origin, so finding every copy of one plan
    /// costs the items that carry its origin and never a page of the board.
    async fn copied_from(
        &self,
        origin: &str,
        kind: BoardKind,
    ) -> Result<Vec<Resolved>, SourceError> {
        let wanted = TaskQuery {
            origin: Some(origin.to_owned()),
            ..TaskQuery::default()
        };
        Ok(self
            .narrowed(&wanted)
            .await?
            .unwrap_or_default()
            .into_iter()
            .filter(|item| item.kind == kind && wanted.origin_matches(&item.metadata()))
            .collect())
    }

    /// Every item of this board that may carry `origin` — a superset of those that do — found
    /// by [`graphql::ORIGIN_LOOKUP`] and never by enumerating the board.
    ///
    /// The union of the board's own field filter over the `onetaskgraph.origin` text field —
    /// which reads the field every carrier holds, whichever release wrote it — and the
    /// board-scoped issue search for the same id as a phrase in the body, where this source
    /// mirrors it. The caller adds the third read, this process's own writes. Candidates are
    /// returned unconfirmed; [`task_matches`] compares each one's own origin field with the
    /// query's, exactly.
    ///
    /// Both connections are walked to exhaustion, each from its own cursor. One that has
    /// already ended is sent its last cursor again, which answers an empty page, so the one
    /// document serves every page of either. What the two leave is stated in the module
    /// documentation: a carrier another process added within the last second or two, before
    /// either index has it.
    async fn origin_candidates(&self, origin: &str) -> Result<Vec<Resolved>, SourceError> {
        let filter = format!("{ORIGIN_FIELD}:{}", quoted(origin));
        let search = self.board_search(Some(&format!("in:body {}", quoted(&as_stored(origin)))));
        let mut items_after: Option<String> = None;
        let mut search_after: Option<String> = None;
        let mut found: Vec<Resolved> = Vec::new();
        let keep = |resolved: Resolved, found: &mut Vec<Resolved>| {
            if !found.iter().any(|held| held.id == resolved.id) {
                found.push(resolved);
            }
        };
        loop {
            let data = self
                .graphql(
                    graphql::ORIGIN_LOOKUP,
                    json!({"owner":self.owner,"number":self.project_number,"filter":filter,
                           "search":search,"type":"ISSUE","originFirst":ORIGIN_PAGE_SIZE,
                           "itemsAfter":items_after,"searchAfter":search_after,
                           "nestedFirst":NESTED_PAGE_SIZE,"boardItems":BOARD_ITEMS_PAGE_SIZE,
                           "duplicates":true}),
                )
                .await?;
            let items = data
                .pointer("/originItems/projectV2/items")
                .filter(|value| !value.is_null())
                .ok_or_else(|| SourceError::Refused {
                    message: format!(
                        "GitHub project {}/{} was not found or is not visible to the token",
                        self.owner, self.project_number
                    ),
                })?;
            for item in optional_nodes(Some(items), "project items")?
                .into_iter()
                .flatten()
            {
                // The board's own items list its drafts too, and a draft is not an issue: no
                // narrowed read answers with one, whatever its origin field holds.
                if let Some(resolved) = self.resolve(item)?
                    && resolved.content_kind == ContentKind::Issue
                {
                    keep(resolved, &mut found);
                }
            }
            let searched = data.get("search").ok_or_else(|| SourceError::Malformed {
                message: "GitHub search response has no search connection".into(),
            })?;
            for node in optional_nodes(Some(searched), "search")?
                .into_iter()
                .flatten()
            {
                if let Some(resolved) = self.resolve_issue(node).await? {
                    keep(resolved, &mut found);
                }
            }
            let items_next = resumed(items, items_after.as_deref())?;
            let search_next = resumed(searched, search_after.as_deref())?;
            if !items_next.has_more() && !search_next.has_more() {
                return Ok(found);
            }
            items_after = items_next.cursor();
            search_after = search_next.cursor();
        }
    }

    /// `found`, with every item this process created or wrote in its place, and every one of
    /// them the read did not report added.
    ///
    /// This process's own record wins over the read's copy of the same item, because a read
    /// of an item written moments ago can still be behind what was written onto it — the
    /// origin field included, which is the one a narrowed read is confirmed against — and a
    /// read that still names an item under a predicate this process's write moved it out of
    /// must not return it. The one thing the read knows that the record cannot is when GitHub
    /// last saw the item change, which is what a comment-activity read rules a candidate out
    /// by, so the read's `updatedAt` is kept when the record has none of its own. See
    /// [`Self::created`] and [`Self::updated`](GitHubProjectsSource::updated).
    fn with_own_writes(&self, mut found: Vec<Resolved>) -> Result<Vec<Resolved>, SourceError> {
        // A board draft is not an issue, so no narrowed read returns one, and this process
        // having written one does not make it an answer either.
        let own: Vec<Resolved> = self
            .created()?
            .iter()
            .chain(self.updated()?.iter())
            .filter(|own| own.content_kind == ContentKind::Issue)
            .cloned()
            .collect();
        for mut own in own {
            self.resolved_cache()?.insert(own.id.clone(), own.clone());
            match found.iter_mut().find(|read| read.id == own.id) {
                Some(read) => {
                    own.updated_at = own.updated_at.max(read.updated_at);
                    *read = own;
                }
                None => found.push(own),
            }
        }
        Ok(found)
    }

    /// Whether `item` has a comment created or last edited at or after `since` — always, when
    /// there is no instant to hold it to.
    ///
    /// The candidate's own `updatedAt` is read first, because a comment written or edited at
    /// or after the instant moved it there: an issue not updated since holds no such comment,
    /// and its comments are never asked for — unless this process commented on it in this
    /// command, when the `updatedAt` held may predate that comment; see [`Self::commented`]. Otherwise its comments are walked, oldest first,
    /// only as far as the first that matches. A board draft is not an issue and has no
    /// comments, so it never matches.
    async fn commented_since(
        &self,
        item: &Resolved,
        since: Option<DateTime<Utc>>,
    ) -> Result<bool, SourceError> {
        let Some(since) = since else {
            return Ok(true);
        };
        if item.content_kind == ContentKind::DraftIssue {
            return Ok(false);
        }
        // An `updatedAt` this process's own record or a lagging index holds can predate a
        // comment this process wrote since, so only an issue it did not comment on is ruled
        // out by one.
        if item.updated_at.is_some_and(|updated| updated < since)
            && !self.commented()?.contains(&item.id)
        {
            return Ok(false);
        }
        let query = TaskQuery {
            commented_since: Some(since),
            ..TaskQuery::default()
        };
        let mut after: Option<String> = None;
        // A first page read beside its fellow candidates answers as this read's own would.
        let held = self.held_comments()?.get(&item.id).cloned();
        if let Some(first) = held {
            if query.comments_match(&first.items) {
                return Ok(true);
            }
            match first.next {
                Some(next) => after = Some(next.0),
                None => return Ok(false),
            }
        }
        loop {
            let data = self
                .graphql(
                    graphql::ISSUE_COMMENTS,
                    json!({"id":item.id.0,"first":MAX_PAGE_SIZE,"after":after}),
                )
                .await?;
            let Some(connection) = data
                .get("node")
                .filter(|value| !value.is_null())
                .and_then(|node| node.get("comments"))
                .filter(|value| !value.is_null())
            else {
                // Removed since the search reported it: no longer an issue with comments.
                return Ok(false);
            };
            let comments = optional_nodes(Some(connection), "issue comments")?
                .into_iter()
                .flatten()
                .map(comment_from)
                .collect::<Result<Vec<_>, _>>()?;
            if query.comments_match(&comments) {
                return Ok(true);
            }
            match next_cursor(connection)? {
                Some(next) => {
                    validate_cursor_progress(after.as_deref(), &next.0)?;
                    after = Some(next.0);
                }
                None => return Ok(false),
            }
        }
    }

    /// Every item on the board: the union of both enumerations GitHub offers of one.
    ///
    /// Neither contains the other, so neither is dropped — only `ProjectV2.items` lists a
    /// board **draft** and reads the board's own fields beside its items, and only the search
    /// reports an item that connection is behind on. The module documentation is where the lag and the
    /// measurements behind it are written down.
    ///
    /// A search result is admitted on the same terms as any other issue this source reaches
    /// directly — [`Self::resolve_issue`] keeps it only if that issue's own `projectItems`
    /// names *this* board — so an issue the index still believes is here after it was taken
    /// off is refused rather than reported.
    ///
    /// See [`Self::board_cache`]. Both completions happen on every call rather than once,
    /// which is what the cache could otherwise have broken.
    async fn board(&self) -> Result<Board, SourceError> {
        self.board_completed_by(None).await
    }

    /// [`Self::board`], its search half narrowed by `also` when it is given — which an issue
    /// that half could add is then answered only if it holds.
    async fn board_completed_by(&self, also: Option<&str>) -> Result<Board, SourceError> {
        let cached = self.board_cache()?.clone();
        let mut board = match cached {
            Some(board) => board,
            None => {
                let read = self.read_board().await?;
                *self.board_cache()? = Some(read.clone());
                read
            }
        };
        let searched = match also {
            None => self.searched_issues().await?,
            Some(also) => self.searched_narrowly(also).await?,
        };
        // A complete sweep asks for every page of this view. Reconcile ids once per
        // view, rather than scanning the board for every search result on every page.
        let mut ids: std::collections::HashSet<_> =
            board.items.iter().map(|item| item.id.clone()).collect();
        for held in searched {
            if ids.insert(held.id.clone()) {
                board.items.push(held);
            }
        }
        for own in self.created()?.iter() {
            if ids.insert(own.id.clone()) {
                board.items.push(own.clone());
            }
        }
        Ok(board)
    }

    /// This process's own view of the board, or the refusal a poisoned lock is.
    fn board_cache(&self) -> Result<std::sync::MutexGuard<'_, Option<Board>>, SourceError> {
        self.board_cache
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "this source's view of the board was left inconsistent by an earlier \
                      failure; next: run the command again"
                    .into(),
            })
    }

    /// Bring this process's own view of the board up to an item it has just written.
    ///
    /// A created item goes to `created`, which is what completes a board read GitHub's own
    /// eventual consistency has left behind. An item that was already there is replaced
    /// where it sits, so a second write of it in the same command reads its real parent
    /// rather than the one it had before the first write.
    ///
    /// "Where it sits" is three places, and missing an earlier one leaves a stale record
    /// that wins: an item this same run created is held in `created` and not in the cached
    /// board, and `board` completes the cached board *from* `created`, so replacing only
    /// the cached copy of such an item replaces nothing and the read still reports the
    /// title it was created with. The search is the third, and it is the one an item the
    /// board's own projection is behind on sits in *alone* — which is exactly the item this
    /// source is least able to re-read, so leaving it out would put the stale title back on
    /// the only items the completion in [`Self::board`] exists for.
    fn remember_written(&self, item: Resolved, created: bool) -> Result<(), SourceError> {
        self.resolved_cache()?.insert(item.id.clone(), item.clone());
        if created {
            self.created()?.push(item);
            return Ok(());
        }
        {
            let mut own = self.created()?;
            if let Some(held) = own.iter_mut().find(|held| held.id == item.id) {
                *held = item;
                return Ok(());
            }
        }
        {
            let mut own = self.updated()?;
            match own.iter_mut().find(|held| held.id == item.id) {
                Some(held) => *held = item.clone(),
                None => own.push(item.clone()),
            }
        }
        if let Some(board) = self.board_cache()?.as_mut()
            && let Some(held) = board.items.iter_mut().find(|held| held.id == item.id)
        {
            *held = item.clone();
        }
        if let Some(found) = self.search_cache()?.as_mut()
            && let Some(held) = found.iter_mut().find(|held| held.id == item.id)
        {
            *held = item.clone();
        }
        for found in self.narrowed_cache()?.values_mut() {
            if let Some(held) = found.iter_mut().find(|held| held.id == item.id) {
                *held = item.clone();
            }
        }
        for (_, found) in self.children_cache()?.values_mut() {
            if let Some(held) = found.iter_mut().find(|held| held.id == item.id) {
                *held = item.clone();
            }
        }
        Ok(())
    }

    /// Forget one item this process has just deleted, from every half of its own view.
    fn forget(&self, id: &NativeId) -> Result<(), SourceError> {
        if let Some(Deferred::Held { created, .. }) = self.deferred()?.as_mut() {
            created.retain(|held| held.content != *id);
        }
        self.resolved_cache()?.remove(id);
        self.read_ahead()?.remove(id);
        self.carried()?.remove(id);
        self.created()?.retain(|own| own.id != *id);
        self.updated()?.retain(|own| own.id != *id);
        self.commented()?.retain(|own| own != id);
        if let Some(board) = self.board_cache()?.as_mut() {
            board.items.retain(|item| item.id != *id);
        }
        if let Some(found) = self.search_cache()?.as_mut() {
            found.retain(|item| item.id != *id);
        }
        for found in self.narrowed_cache()?.values_mut() {
            found.retain(|item| item.id != *id);
        }
        for (_, found) in self.children_cache()?.values_mut() {
            found.retain(|item| item.id != *id);
        }
        Ok(())
    }

    /// This process's own record of each narrowed answer, or the refusal a poisoned lock is.
    fn narrowed_cache(
        &self,
    ) -> Result<std::sync::MutexGuard<'_, BTreeMap<String, Vec<Resolved>>>, SourceError> {
        self.narrowed_cache
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "this source's view of a narrowed read was left inconsistent by an \
                      earlier failure; next: run the command again"
                    .into(),
            })
    }

    /// This process's own record of each project's sub-issues, or the refusal a poisoned lock
    /// is.
    /// The tasks `project` holds, as a copy's batched read found them this command, when it did.
    fn held_task_children(&self, project: &NativeId) -> Result<Option<Vec<Resolved>>, SourceError> {
        Ok(self.task_children()?.get(project).cloned())
    }

    fn task_children(
        &self,
    ) -> Result<std::sync::MutexGuard<'_, BTreeMap<NativeId, Vec<Resolved>>>, SourceError> {
        self.task_children
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "this source's record of a project's tasks was left inconsistent by an \
                          earlier failure; next: run the command again"
                    .into(),
            })
    }

    fn children_cache(&self) -> Result<std::sync::MutexGuard<'_, ProjectChildren>, SourceError> {
        self.children_cache
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "this source's view of a project's tasks was left inconsistent by an \
                      earlier failure; next: run the command again"
                    .into(),
            })
    }

    /// Every page of the board, read from GitHub.
    async fn read_board(&self) -> Result<Board, SourceError> {
        let mut after: Option<String> = None;
        let mut items = Vec::new();
        let mut board;
        loop {
            let page = self.board_page(after.as_deref(), MAX_PAGE_SIZE).await?;
            for item in page
                .pointer("/items/nodes")
                .and_then(Value::as_array)
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub project items.nodes is not an array".into(),
                })?
            {
                if let Some(resolved) = self.resolve(item)? {
                    items.push(resolved);
                }
            }
            let info = page
                .pointer("/items/pageInfo")
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub project items have no pageInfo".into(),
                })?;
            let has_next = required_bool(info, "hasNextPage")?;
            let next = has_next
                .then(|| required_str(info, "endCursor"))
                .transpose()?;
            board = page.clone();
            match next {
                Some(next) => {
                    validate_cursor_progress(after.as_deref(), next)?;
                    after = Some(next.to_owned());
                }
                None => break,
            }
        }
        Ok(Board {
            id: required_str(&board, "id")?.to_owned(),
            fields: board.get("fields").cloned().unwrap_or(Value::Null),
            items,
        })
    }

    /// The existing items this source has written, for completing a narrowed read that is
    /// behind; see [`Self::updated`](GitHubProjectsSource::updated).
    fn updated(&self) -> Result<std::sync::MutexGuard<'_, Vec<Resolved>>, SourceError> {
        self.updated.lock().map_err(|_| SourceError::Unavailable {
            message: "this source's record of what it wrote in this run was left inconsistent \
                      by an earlier failure; next: run the command again"
                .into(),
        })
    }

    /// The issues this source has commented on in this command; see
    /// [`Self::commented`](GitHubProjectsSource::commented).
    fn commented(&self) -> Result<std::sync::MutexGuard<'_, Vec<NativeId>>, SourceError> {
        self.commented.lock().map_err(|_| SourceError::Unavailable {
            message: "this source's record of what it commented on in this run was left \
                      inconsistent by an earlier failure; next: run the command again"
                .into(),
        })
    }

    /// Called only once GitHub has answered the comment write, so an issue whose comment
    /// failed is never made a candidate a later read would pay a node read for.
    fn remember_commented(&self, issue: &NativeId) -> Result<(), SourceError> {
        // A comment page held for this command is no longer what the issue holds.
        self.held_comments()?.remove(issue);
        let mut commented = self.commented()?;
        if !commented.contains(issue) {
            commented.push(issue.clone());
        }
        Ok(())
    }

    /// The items this source has created, for completing a board read that is behind.
    fn created(&self) -> Result<std::sync::MutexGuard<'_, Vec<Resolved>>, SourceError> {
        self.created.lock().map_err(|_| SourceError::Unavailable {
            message: "this source's record of what it created in this run was left \
                      inconsistent by an earlier failure; next: run the command again"
                .into(),
        })
    }

    /// One board item as this source reports it, or `None` for content it ignores.
    ///
    /// A pull request is neither a project nor a task — it is somebody's change, not a
    /// unit of plan — and an item whose content the token cannot see has nothing to
    /// report at all.
    fn resolve(&self, item: &Value) -> Result<Option<Resolved>, SourceError> {
        let resolved = self.resolve_unremembered(item)?;
        if let Some(resolved) = &resolved {
            self.resolved_cache()?
                .insert(resolved.id.clone(), resolved.clone());
        }
        Ok(resolved)
    }

    /// [`Self::resolve`], without remembering the item it resolves: what reading a token off a
    /// write's own response needs, which says nothing this process should answer a later read
    /// with.
    fn resolve_unremembered(&self, item: &Value) -> Result<Option<Resolved>, SourceError> {
        let content = item.get("content").ok_or_else(|| SourceError::Malformed {
            message: "GitHub project item is missing content".into(),
        })?;
        if content.is_null() {
            return Ok(None);
        }
        let content_kind = match required_str(content, "__typename")? {
            "Issue" => ContentKind::Issue,
            "DraftIssue" => ContentKind::DraftIssue,
            _ => return Ok(None),
        };
        let field_values = item
            .get("fieldValues")
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub project item is missing fieldValues".into(),
            })?;
        complete_connection(field_values, "project item field values", NESTED_PAGE_SIZE)?;
        let nodes = field_values
            .get("nodes")
            .and_then(Value::as_array)
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub project item fieldValues.nodes is not an array".into(),
            })?;
        if let Some(labels) = content.get("labels") {
            complete_connection(labels, "content labels", NESTED_PAGE_SIZE)?;
        }
        let raw_body = optional_str(content, "body")?.map(str::to_owned);
        let (body, slot) = self.visible(raw_body.clone())?;
        let parent = optional_str(content.get("parent").unwrap_or(&Value::Null), "id")?
            .map(|id| NativeId(id.to_owned()));
        // A draft has no sub-issues to summarise, and GitHub's schema gives it no field
        // to read one from; it is a task, and never a project.
        let sub_issues = match content_kind {
            ContentKind::Issue => sub_issue_total(content)?,
            ContentKind::DraftIssue => 0,
        };
        let content_id = required_str(content, "id")?;
        let marked = ItemKind::from_metadata(&slot).map_err(|message| SourceError::Malformed {
            message: format!("GitHub issue {content_id}: {message}"),
        })?;
        let raw_title = required_str(content, "title")?;
        // Read as children, a sub-issue is filed under no project and nothing is a project: the
        // GitHub parent is the parent relation instead.
        let (kind, parent, issue_parent) = match self.sub_issues {
            SubIssues::Project => (
                BoardKind::of(raw_title, parent.is_some(), sub_issues, marked),
                parent,
                None,
            ),
            SubIssues::Parent => (
                if raw_title.starts_with(DESIGN_TITLE_PREFIX) {
                    BoardKind::Document
                } else {
                    BoardKind::Work(ItemKind::Task)
                },
                None,
                parent,
            ),
        };
        // The title a person wrote, which for a document is the one without the prefix —
        // the same way `content` above is the body without this source's metadata slot.
        let title = match kind {
            BoardKind::Document => raw_title[DESIGN_TITLE_PREFIX.len()..].to_owned(),
            BoardKind::Work(_) => raw_title.to_owned(),
        };
        let own_repository = content
            .pointer("/repository/nameWithOwner")
            .and_then(Value::as_str)
            .map(|origin| Repository::try_from(format!("{}/{origin}", RepositoryTarget::HOST)))
            .transpose()
            .map_err(|message| SourceError::Malformed { message })?;
        let repositories = if slot.contains_key(Repository::METADATA_KEY) {
            Repository::from_metadata(&slot)
                .map_err(|message| SourceError::Malformed { message })?
        } else {
            own_repository.clone().into_iter().collect()
        };
        let classification = Classification::from_metadata(&slot)
            .map_err(|message| SourceError::Malformed { message })?;
        let id = NativeId(content_id.to_owned());
        // Read only for a task, because only a task has either list: a project or a
        // document holding one of these keys holds nothing this source reports, and the
        // keys are left out of its caller-visible metadata all the same.
        let (delivers, delivered_by) = if kind == BoardKind::Work(ItemKind::Task) {
            let listed = |key: &str| {
                TaskRef::from_value(key, &id, Some(&self.name), slot.get(key))
                    .map_err(|message| SourceError::Malformed { message })
            };
            (
                listed(TaskRef::DELIVERS_KEY)?,
                listed(TaskRef::DELIVERED_BY_KEY)?,
            )
        } else {
            (Vec::new(), Vec::new())
        };
        let (option, closed, reason) = Self::status_parts(nodes, content)?;
        let priority = self.held_priority(nodes)?;
        let item_labels = labels(content)?;
        // Present when the item was reached through its own issue, whose board entry
        // names the board; a read of the board's own items has the board already. An
        // empty id names nothing a field write could address, so it is read as absent and
        // the write goes back to reading the board.
        let board_id = item
            .pointer("/project/id")
            .and_then(Value::as_str)
            .filter(|id| !id.is_empty());
        let resolved = Resolved {
            item_id: required_str(item, "id")?.to_owned(),
            id,
            content_kind,
            kind,
            title,
            body: body.filter(|value| !value.is_empty()),
            raw_body,
            status: self
                .statuses
                .status(kind.status_kind(), option, closed, reason),
            option: option.map(str::to_owned),
            priority,
            closed,
            reason: closed.then(|| reason.map(str::to_owned)).flatten(),
            delivers,
            delivered_by,
            labels: item_labels.clone(),
            parent,
            issue_parent,
            origin: text_field(nodes, ORIGIN_FIELD)?.filter(|value| !value.is_empty()),
            projected: self.held_projections(nodes)?,
            number: match content_kind {
                ContentKind::Issue => Some(issue_number(content)?),
                // A draft is filed in no repository, so nothing ever numbered it:
                // `DraftIssue` declares no `number` at all, exactly as it declares no
                // `subIssuesSummary` the branch above reads.
                ContentKind::DraftIssue => None,
            },
            url: optional_str(content, "url")?.map(str::to_owned),
            created_at: optional_time(content, "createdAt")?,
            updated_at: optional_time(content, "updatedAt")?,
            item_updated_at: optional_time(item, "updatedAt")?,
            comment_count: content
                .pointer("/commentCount/totalCount")
                .and_then(Value::as_u64),
            relationship_events: content
                .pointer("/relationshipEvents/totalCount")
                .and_then(Value::as_u64),
            // The issue node is stamped wherever it was read; a board item a read of the issue
            // put together from its memberships is not.
            served_at: match optional_time(content, SERVED_AT)? {
                Some(served) => Some(served),
                None => optional_time(item, SERVED_AT)?,
            },
            // Both connections were read to their end above — one reporting a further page is
            // refused — so these are every field value and every label the item holds. A
            // response that did not select the labels at all says nothing of them, and gives
            // no facets.
            facets: content.get("labels").is_some().then(|| {
                change_token::Facets::new(kind.facet(), facet_fields(nodes), &item_labels).digest()
            }),
            own_repository,
            repositories,
            classification,
            slot,
            board_id: board_id.map(str::to_owned),
            fields: field_definitions(nodes),
            board_fields: Self::carried_board_fields(content, board_id)?,
            blocked_by: carried_blocked_by(content)?,
            assignees: if self.assignees {
                Some(assignees::assignee_logins(content)?)
            } else {
                None
            },
        };
        Ok(Some(resolved))
    }

    /// The field definitions of the board `board_id` names — the project this issue's own
    /// board item is on — off the `boards` page a read of an issue by its own id carries, or
    /// `None` when the read carried none, carried no entry for that board, or the board item
    /// named no board, which a write then answers by reading the board's fields itself.
    ///
    /// Matched by the board's node id and never by its number alone: a project number is
    /// unique only within its owner, so another owner's board numbered alike can sit on the
    /// same page, and its field and option ids address nothing on this one.
    fn carried_board_fields(
        content: &Value,
        board_id: Option<&str>,
    ) -> Result<Option<Value>, SourceError> {
        let (Some(nodes), Some(board_id)) = (
            content.pointer("/boards/nodes").and_then(Value::as_array),
            board_id,
        ) else {
            return Ok(None);
        };
        let Some(board) = nodes.iter().find_map(|node| {
            let project = node.get("project")?;
            (project.get("id").and_then(Value::as_str) == Some(board_id)).then_some(project)
        }) else {
            return Ok(None);
        };
        let Some(fields) = board.get("fields").filter(|fields| !fields.is_null()) else {
            return Ok(None);
        };
        complete_connection(fields, "board fields", NESTED_PAGE_SIZE)?;
        Ok(Some(fields.clone()))
    }

    /// What one board item's `Priority` field says, through this instance's mapping.
    ///
    /// An instance with no mapping holds no priority, so every item reads as `none` whatever
    /// its board holds. With one, no value is `none`, a mapped option is its level, and an
    /// option the mapping does not name is kept as itself — never read as a level or as
    /// `none` — for a read of the task to report by name.
    fn held_priority(&self, field_values: &[Value]) -> Result<HeldPriority, SourceError> {
        let Some(mapping) = &self.priorities else {
            return Ok(HeldPriority::Read(Priority::None));
        };
        // A value of the field that names no option — a text field someone called `Priority` —
        // is malformed rather than `none`: reading it as no priority would let the next copy
        // clear one a person set.
        let Some(option) = field_values
            .iter()
            .find(|value| {
                value.pointer("/field/name").and_then(Value::as_str) == Some(PRIORITY_FIELD)
            })
            .map(|value| required_str(value, "name"))
            .transpose()?
        else {
            return Ok(HeldPriority::Read(Priority::None));
        };
        Ok(mapping.priority_of(option).map_or_else(
            || HeldPriority::Unmapped(option.to_owned()),
            HeldPriority::Read,
        ))
    }

    /// What each projected metadata field holds for one board item, by field name — every text
    /// value GitHub answers, an empty one included.
    fn held_projections(
        &self,
        field_values: &[Value],
    ) -> Result<BTreeMap<String, String>, SourceError> {
        let mut held = BTreeMap::new();
        for projection in &self.metadata_fields {
            if let Some(text) = text_field(field_values, &projection.field)? {
                held.insert(projection.field.to_string(), text);
            }
        }
        Ok(held)
    }

    /// What one board item's status is read from: its `Status` option, whether its issue
    /// is closed, and the reason it was closed with. [`BoardStatuses::status`] turns the
    /// three into the status it reports.
    fn status_parts<'a>(
        field_values: &'a [Value],
        content: &'a Value,
    ) -> Result<(Option<&'a str>, bool, Option<&'a str>), SourceError> {
        let option = field_values
            .iter()
            .find(|value| value.pointer("/field/name").and_then(Value::as_str) == Some("Status"))
            .map(|value| required_str(value, "name"))
            .transpose()?;
        let closed = optional_str(content, "state")? == Some("CLOSED");
        Ok((option, closed, optional_str(content, "stateReason")?))
    }

    /// The board Status option this write selects, or the refusal that says why not.
    ///
    /// The mapped option is required for both open and terminal targets. A terminal write
    /// validates it before changing either representation, so it can never fall back to
    /// closing an issue whose board cannot display the matching status.
    ///
    /// Answers the field's id, the option's id, and the option's name as the board spells
    /// it — which is the name a read of the item reports once it sits there.
    fn column_for(
        &self,
        fields: &Value,
        kind: ItemKind,
        category: StatusCategory,
        target: &StatusTarget,
    ) -> Result<Option<(String, String, String)>, SourceError> {
        let Some(wanted) = target.option() else {
            return Ok(None);
        };
        let missing = |detail: &str| SourceError::Refused {
            message: format!(
                "{} status {} of source {} needs the board Status option {wanted:?}, and \
                 {detail}; next: add that option to the board, which `onetaskgraph sources \
                 fields {} --apply` does, or point status_mapping.{}.{} of this source at one \
                 it has",
                kind.marker(),
                category_name(category),
                self.name,
                self.name,
                category_name(category),
                kind.marker()
            ),
        };
        let Some(field) = Board::field(fields, "Status")? else {
            return Err(missing("this board has no Status field"));
        };
        if required_str(field, "__typename")? != "ProjectV2SingleSelectField" {
            return Err(missing(
                "this board's Status field is not a single-select field",
            ));
        }
        let option = field
            .get("options")
            .and_then(Value::as_array)
            .and_then(|options| {
                options.iter().find(|option| {
                    option
                        .get("name")
                        .and_then(Value::as_str)
                        .is_some_and(|name| name.eq_ignore_ascii_case(wanted))
                })
            });
        match option {
            None => Err(missing("this board does not have it")),
            Some(option) => Ok(Some((
                required_str(field, "id")?.to_owned(),
                required_str(option, "id")?.to_owned(),
                required_str(option, "name")?.to_owned(),
            ))),
        }
    }

    /// The refusal a status that closes an issue is answered with over a board draft.
    fn closes_a_draft(&self, category: StatusCategory) -> SourceError {
        SourceError::Refused {
            message: format!(
                "status {} of source {} closes the item's issue, and GitHub draft items have \
                 no open or closed state",
                category_name(category),
                self.name
            ),
        }
    }

    /// What a status write to one item needs of the board: the board's id and the
    /// definition of its `Status` field, read off the item when the item says both.
    ///
    /// The same reasoning as [`Self::fields_for`]: a node read of the item names its board,
    /// and its `Status` value carries that field's definition, options and all. An item that
    /// does not say — no board id, or no `Status` value to read the field off — takes them
    /// from [`Self::board_fields`], which reads no item.
    async fn status_board(&self, item: &Resolved) -> Result<BoardFields, SourceError> {
        if let Some(board) = item.carried_board() {
            return Ok(board);
        }
        if item.defines("Status")
            && let Some(board_id) = item.named_board()
        {
            return Ok(BoardFields {
                id: board_id,
                fields: json!({"nodes": item.fields, "pageInfo": {"hasNextPage": false}}),
            });
        }
        self.board_fields().await
    }

    /// Set one task's or one project's status and nothing else; see
    /// [`TaskSource::set_task_status`] and [`TaskSource::set_project_status`].
    async fn set_status(
        &self,
        id: &NativeId,
        kind: ItemKind,
        category: StatusCategory,
    ) -> Result<Option<Status>, SourceError> {
        // Refused before anything is read, in the words a write of the same status is.
        let target = self.resolved_target(kind, category)?;
        // An item read by issue reference carries only what this write needs, so what it
        // becomes is never remembered for a later read to be answered from.
        let held = self.take_held(id)?;
        let by_reference = held.is_some();
        let bound = match held {
            Some(item) => Some(item),
            None => self.bound_item(id).await?,
        };
        let Some(mut item) = bound.filter(|item| item.kind == BoardKind::Work(kind)) else {
            return Ok(None);
        };
        let board = self.status_board(&item).await?;
        let (field, option, name) = self
            .column_for(&board.fields, kind, category, &target)?
            .ok_or_else(|| SourceError::Malformed {
                message: format!(
                    "status {} of source {} names no board Status option",
                    category_name(category),
                    self.name
                ),
            })?;
        if item.status.category == category && item.option.as_deref() == Some(&name) {
            return Ok(Some(item.status));
        }
        match &target {
            StatusTarget::Terminal(_, reason) => {
                if item.content_kind == ContentKind::DraftIssue {
                    return Err(self.closes_a_draft(category));
                }
                // The option first, so a refused selection leaves an open issue open rather
                // than closed behind a card that still reads open.
                self.set_item_field(
                    board.id.as_str(),
                    &item.item_id,
                    &field,
                    json!({"singleSelectOptionId": option}),
                )
                .await?;
                // An issue already closed in a way that reads as this category keeps its state
                // and its reason: only its card was out of line, and closing it again would
                // overwrite a reason — a duplicate's — that says more than this write does.
                let already = item.closed && item.status.category == category;
                if !already {
                    self.update_content(
                        ContentKind::Issue,
                        &item.id,
                        json!({"stateInput": state_input(Some(&target))}),
                    )
                    .await?;
                }
                let reason = if already {
                    item.reason.clone()
                } else {
                    Some(reason.reason().to_owned())
                };
                item.closed = true;
                item.status = self
                    .statuses
                    .status(kind, Some(&name), true, reason.as_deref());
                item.option = Some(name);
                item.reason = reason;
            }
            StatusTarget::Column(_) => {
                // An option is what an open item's status is, so a closed issue is reopened
                // first — sitting closed in the column, it would read back as closed. A draft has
                // no state to reopen.
                if item.content_kind == ContentKind::Issue && item.closed {
                    self.update_content(
                        ContentKind::Issue,
                        &item.id,
                        json!({"stateInput": state_input(Some(&target))}),
                    )
                    .await?;
                    item.closed = false;
                }
                self.set_item_field(
                    board.id.as_str(),
                    &item.item_id,
                    &field,
                    json!({"singleSelectOptionId": option}),
                )
                .await?;
                item.status = self.statuses.status(kind, Some(&name), false, None);
                item.option = Some(name);
            }
            StatusTarget::Disabled(_) => {
                unreachable!("resolved_target refused a disabled status")
            }
        }
        let status = item.status.clone();
        if !by_reference {
            self.remember_written(item, false)?;
        }
        Ok(Some(status))
    }

    /// Replace one task's `delivered_by` and nothing else; see
    /// [`TaskSource::set_delivered_by`].
    ///
    /// One update of the body, which differs from the body GitHub holds only inside the
    /// metadata slot — see [`with_slot`]. A body that would not change is not sent at all.
    async fn replace_delivered_by(
        &self,
        id: &NativeId,
        delivered_by: &[TaskRef],
    ) -> Result<Option<()>, SourceError> {
        let entries = TaskRef::listed(
            TaskRef::DELIVERED_BY_KEY,
            id,
            Some(&self.name),
            delivered_by.to_vec(),
        )
        .map_err(|message| SourceError::Refused { message })?;
        let Some(mut item) = self
            .bound_item(id)
            .await?
            .filter(|item| item.kind == BoardKind::Work(ItemKind::Task))
        else {
            return Ok(None);
        };
        let mut slot = item.slot.clone();
        set_task_list(&mut slot, TaskRef::DELIVERED_BY_KEY, &entries);
        self.write_slot(&mut item, &slot).await?;
        item.delivered_by = entries;
        self.remember_written(item, false)?;
        Ok(Some(()))
    }

    /// Set one caller key of the metadata slot of one issue of `kind`, and nothing else;
    /// see [`TaskSource::set_task_metadata`].
    ///
    /// `None` when this board holds no item by that id, or holds one of another kind. The
    /// answer is the item as this source now reads it, so what a caller is told the key
    /// holds is what the slot holds.
    ///
    /// A key already holding the value is answered without a write, compared as JSON rather
    /// than as the body's bytes: a slot a person spelled with other whitespace would
    /// otherwise be re-encoded, which is a write that changes nothing the caller asked for.
    async fn set_slot_key(
        &self,
        id: &NativeId,
        kind: BoardKind,
        key: &MetadataKey,
        value: &Value,
    ) -> Result<Option<Resolved>, SourceError> {
        let Some(mut item) = self.bound_item(id).await?.filter(|item| item.kind == kind) else {
            return Ok(None);
        };
        if item.slot.get(key.as_str()) == Some(value) {
            return Ok(Some(item));
        }
        let mut slot = item.slot.clone();
        slot.insert(key.as_str().to_owned(), value.clone());
        // A key a board text field projects moves that field first, so the body — the value's
        // one authoritative home — is written last, as every write of an existing item is.
        if self
            .metadata_fields
            .iter()
            .any(|projection| &*projection.key == key.as_str())
        {
            let board = self.projection_board(&item).await?;
            let (writes, projected) = self.projection_writes(
                &board.fields,
                Some(&item.projected),
                &slot,
                Some(&[key.as_str()]),
            )?;
            let (mut values, mut clears) = (Vec::new(), Vec::new());
            for write in &writes {
                write.join(&mut values, &mut clears);
            }
            self.set_item_fields(board.id.as_str(), &item.item_id, &values, &clears)
                .await?;
            item.projected = projected;
        }
        self.write_slot(&mut item, &slot).await?;
        self.remember_written(item.clone(), false)?;
        Ok(Some(item))
    }

    /// What a projection write to `item` needs of the board, read off the item when it says
    /// enough, as [`Self::fields_for`] reads it.
    async fn projection_board(&self, item: &Resolved) -> Result<BoardFields, SourceError> {
        if let Some(board) = item.carried_board() {
            return Ok(board);
        }
        if let Some(board_id) = item.named_board()
            && self
                .metadata_fields
                .iter()
                .all(|projection| item.defines(&projection.field))
        {
            return Ok(BoardFields {
                id: board_id,
                fields: json!({"nodes": item.fields, "pageInfo": {"hasNextPage": false}}),
            });
        }
        self.board_fields().await
    }

    /// Put `slot` in one item's metadata slot with a single update of its body, and bring
    /// `item` up to what that write left.
    ///
    /// The body sent differs from the body GitHub holds only inside the slot — see
    /// [`with_slot`] — and a body that would not change is not sent at all. It goes through
    /// the mutation the item's content takes, so a board draft's body is written with
    /// `updateProjectV2DraftIssue` exactly as an issue's is with `updateIssue`.
    async fn write_slot(
        &self,
        item: &mut Resolved,
        slot: &BTreeMap<String, Value>,
    ) -> Result<(), SourceError> {
        let held = item.raw_body.clone().unwrap_or_default();
        let body = with_slot(&held, slot)?;
        if body != held {
            self.update_content(item.content_kind, &item.id, json!({"body": body}))
                .await?;
        }
        let (visible, slot) = self.visible(Some(body.clone()))?;
        item.body = visible.filter(|value| !value.is_empty());
        item.raw_body = Some(body);
        item.slot = slot;
        Ok(())
    }

    /// This instance's target for a category written to an item of `kind`, refusing one
    /// that kind has no option for — before anything is read or written.
    ///
    /// Nothing here mutates the board's option set to make room for a status. GitHub
    /// documents `UpdateProjectV2FieldInput.singleSelectOptions` as *"provided values
    /// overwrite existing options"*, so no addition is additive and a mistake destroys the
    /// field and every item's status.
    fn resolved_target(
        &self,
        kind: ItemKind,
        category: StatusCategory,
    ) -> Result<StatusTarget, SourceError> {
        let target = self.statuses.target(kind, category).clone();
        let StatusTarget::Disabled(why) = target else {
            return Ok(target);
        };
        let refusal = why.refusal(&self.name, category, kind);
        // Why there is no shipped default, which is the question a person meeting this
        // refusal on a source that never mentioned the category asks.
        let shipped_none = match category {
            StatusCategory::Draft => Some(
                "draft has no shipped default because GitHub draft issues cannot have \
                 sub-issues, and this source stores a project's tasks as its issue's sub-issues",
            ),
            StatusCategory::Unknown => Some(
                "unknown has no shipped default because this board keeps no open-ended status \
                 word: every word classified unknown is written to the one board Status option \
                 status_mapping.unknown names",
            ),
            _ => None,
        };
        Err(match (refusal, shipped_none, why) {
            (SourceError::Refused { message }, Some(note), UnmappedStatus::Unconfigured) => {
                SourceError::Refused {
                    message: format!("{message}; {note}"),
                }
            }
            (refusal, _, _) => refusal,
        })
    }

    /// What writing `priority` does to one item's `Priority` field on this board, or the
    /// refusal naming what the board lacks.
    ///
    /// `none` is no value, so it clears the field — and asks nothing of an item that holds
    /// none already, or of an item not created yet. Every other priority selects the option
    /// the mapping names, matched case-insensitively; a board with no `Priority` field, or
    /// without that option, is refused rather than given one: reads and writes never create
    /// a field or an option.
    fn priority_write(
        &self,
        fields: &Value,
        existing: Option<&Resolved>,
        priority: Priority,
    ) -> Result<Option<PriorityWrite>, SourceError> {
        let Some(mapping) = &self.priorities else {
            return Err(self.holds_no_priority());
        };
        let Some(wanted) = mapping.option(priority) else {
            if !existing.is_some_and(Resolved::holds_priority) {
                return Ok(None);
            }
            let field =
                Board::field(fields, PRIORITY_FIELD)?.ok_or_else(|| SourceError::Malformed {
                    message: format!(
                        "an item holding a {PRIORITY_FIELD} value was read without that field"
                    ),
                })?;
            return Ok(Some(PriorityWrite::Clear {
                field: required_str(field, "id")?.to_owned(),
            }));
        };
        let missing = |detail: &str| SourceError::Refused {
            message: format!(
                "priority {priority} of source {} needs the board {PRIORITY_FIELD} option \
                 {wanted:?}, and {detail}; run `onetaskgraph sources fields {} --apply` to add \
                 it, or point priority_mapping.{priority} of this source at an option the board \
                 has",
                self.name, self.name
            ),
        };
        let Some(field) = Board::field(fields, PRIORITY_FIELD)? else {
            return Err(missing(&format!(
                "this board has no {PRIORITY_FIELD} field"
            )));
        };
        if required_str(field, "__typename")? != "ProjectV2SingleSelectField" {
            return Err(missing(&format!(
                "this board's {PRIORITY_FIELD} field is not a single-select field"
            )));
        }
        // An options list that is absent or not a list is an answer this source cannot read,
        // not a board lacking the option: `sources fields --apply` is no remedy for it.
        let option = field
            .get("options")
            .and_then(Value::as_array)
            .ok_or_else(|| SourceError::Malformed {
                message: format!("GitHub {PRIORITY_FIELD} field options is not an array"),
            })?
            .iter()
            .find(|option| {
                option
                    .get("name")
                    .and_then(Value::as_str)
                    .is_some_and(|name| name.eq_ignore_ascii_case(wanted))
            })
            .ok_or_else(|| missing("this board does not have it"))?;
        Ok(Some(PriorityWrite::Select {
            field: required_str(field, "id")?.to_owned(),
            option: required_str(option, "id")?.to_owned(),
        }))
    }

    /// What writing `metadata` does to each projected text field of one item, and what those
    /// fields hold after it — or the refusal, before anything is sent.
    ///
    /// `held` is what the item's fields hold now, `None` for an item not created yet. `keys`
    /// narrows the work to the entries projecting one of those metadata keys, for a write that
    /// touches only them; `None` is every entry. A string the field already holds, and nothing
    /// for a field holding nothing, sends nothing. A value no text field can hold is refused,
    /// and so is a board without the field, or with a field of that name that is not a text
    /// field: reads and writes never create a field.
    fn projection_writes(
        &self,
        fields: &Value,
        held: Option<&BTreeMap<String, String>>,
        metadata: &BTreeMap<String, Value>,
        keys: Option<&[&str]>,
    ) -> Result<(Vec<ProjectionWrite>, BTreeMap<String, String>), SourceError> {
        let mut projected = held.cloned().unwrap_or_default();
        let mut writes = Vec::new();
        for projection in self
            .metadata_fields
            .iter()
            .filter(|projection| keys.is_none_or(|keys| keys.contains(&&*projection.key)))
        {
            let wanted = projection.projected(metadata, &self.name)?;
            let field = self.projection_field(fields, projection)?;
            let holding = projected.get(&*projection.field);
            match wanted {
                Projected::Text(text) if holding != Some(&text) => {
                    writes.push(ProjectionWrite::Set {
                        field,
                        text: text.clone(),
                    });
                    projected.insert(projection.field.to_string(), text);
                }
                Projected::Nothing if holding.is_some() => {
                    writes.push(ProjectionWrite::Clear { field });
                    projected.remove(&*projection.field);
                }
                Projected::Text(_) | Projected::Nothing => {}
            }
        }
        Ok((writes, projected))
    }

    /// The id of the board text field one projection writes, or the refusal naming what the
    /// board has instead.
    fn projection_field(
        &self,
        fields: &Value,
        projection: &MetadataField,
    ) -> Result<String, SourceError> {
        let refuse = |detail: String| SourceError::Refused {
            message: format!(
                "source {} projects metadata key {:?} onto the board text field {:?}, and \
                 {detail}; next: run `onetaskgraph sources fields {} --apply` to create it as a \
                 text field, or name another field in metadata_fields",
                self.name, &*projection.key, &*projection.field, self.name
            ),
        };
        let Some(field) = Board::field(fields, &projection.field)? else {
            return Err(refuse("this board has no field of that name".to_owned()));
        };
        let typename = optional_str(field, "__typename")?.unwrap_or("field of another type");
        let data_type = optional_str(field, "dataType")?;
        if typename != "ProjectV2Field" || data_type != Some("TEXT") {
            return Err(refuse(format!(
                "this board's field of that name is not a text field (it is a {}), so it is \
                 left as it is; rename or remove that field first",
                data_type.unwrap_or(typename)
            )));
        }
        Ok(required_str(field, "id")?.to_owned())
    }

    /// Apply one priority write to one board item.
    async fn write_priority(
        &self,
        board_id: &str,
        item_id: &str,
        write: &PriorityWrite,
    ) -> Result<(), SourceError> {
        match write {
            PriorityWrite::Select { field, option } => {
                self.set_item_field(
                    board_id,
                    item_id,
                    field,
                    json!({"singleSelectOptionId": option}),
                )
                .await
            }
            PriorityWrite::Clear { field } => {
                let data = self
                    .graphql(
                        graphql::CLEAR_FIELD,
                        json!({"input":{"projectId":board_id,"itemId":item_id,"fieldId":field},
                            "readPriority":false,"priorityName":PRIORITY_FIELD}),
                    )
                    .await?;
                let returned = data
                    .pointer("/clearProjectV2ItemFieldValue/projectV2Item")
                    .ok_or_else(|| SourceError::Malformed {
                        message: "GitHub field clear returned no project item".into(),
                    })?;
                if required_str(returned, "id")? != item_id {
                    return Err(SourceError::Malformed {
                        message: "GitHub field clear returned the wrong project item".into(),
                    });
                }
                Ok(())
            }
        }
    }

    /// The refusal a priority is answered with by an instance configured with no
    /// `priority_mapping`, which holds none.
    fn holds_no_priority(&self) -> SourceError {
        SourceError::Refused {
            message: format!(
                "source {} holds no task priority: its configuration sets no priority_mapping; \
                 next: set priority_mapping on this source, then run `onetaskgraph sources \
                 fields {} --apply` to set its board up",
                self.name, self.name
            ),
        }
    }

    /// Set one task's priority and nothing else; see [`TaskSource::set_task_priority`].
    ///
    /// One field write — a select, or a clear for `none` — and no title, body, label, state
    /// or `Status` request. Clearing a priority an item does not hold sends nothing.
    async fn set_priority(
        &self,
        id: &NativeId,
        priority: Priority,
    ) -> Result<Option<Priority>, SourceError> {
        if self.priorities.is_none() {
            return Err(self.holds_no_priority());
        }
        let Some(mut item) = self
            .bound_item(id)
            .await?
            .filter(|item| item.kind == BoardKind::Work(ItemKind::Task))
        else {
            return Ok(None);
        };
        if priority == Priority::None && !item.holds_priority() {
            return Ok(Some(priority));
        }
        // The item's own read carries the field's definition whenever it holds a value of
        // it, which a clear always does; a select onto an item holding none reads the board.
        let board = match (item.carried_board(), item.named_board()) {
            (Some(board), _) => board,
            (None, Some(id)) if item.defines(PRIORITY_FIELD) => BoardFields {
                id,
                fields: json!({"nodes": item.fields.clone(), "pageInfo": {"hasNextPage": false}}),
            },
            _ => self.board_fields().await?,
        };
        let Some(write) = self.priority_write(&board.fields, Some(&item), priority)? else {
            return Ok(Some(priority));
        };
        let (document, root, input) = match write {
            PriorityWrite::Select { field, option } => (
                graphql::UPDATE_FIELD,
                "updateProjectV2ItemFieldValue",
                json!({"projectId":board.id.as_str(),"itemId":item.item_id,"fieldId":field,"value":{"singleSelectOptionId":option}}),
            ),
            PriorityWrite::Clear { field } => (
                graphql::CLEAR_FIELD,
                "clearProjectV2ItemFieldValue",
                json!({"projectId":board.id.as_str(),"itemId":item.item_id,"fieldId":field}),
            ),
        };
        let data = self
            .graphql(
                document,
                json!({"input":input,"readPriority":true,"priorityName":PRIORITY_FIELD}),
            )
            .await?;
        let returned = data
            .get(root)
            .and_then(|value| value.get("projectV2Item"))
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub priority write returned no project item".into(),
            })?;
        if required_str(returned, "id")? != item.item_id {
            return Err(SourceError::Malformed {
                message: "GitHub priority write returned the wrong project item".into(),
            });
        }
        let value = returned
            .get("fieldValueByName")
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub priority write returned no priority read-back".into(),
            })?;
        if !value.is_null()
            && value.pointer("/field/name").and_then(Value::as_str) != Some(PRIORITY_FIELD)
        {
            return Err(SourceError::Malformed {
                message: "GitHub priority read-back is not a Priority field value".into(),
            });
        }
        let values = if value.is_null() {
            Vec::new()
        } else {
            vec![value.clone()]
        };
        item.priority = self.held_priority(&values)?;
        let answer = item.task()?.priority;
        self.remember_written(item, false)?;
        Ok(Some(answer))
    }

    /// Replace one task's visible body and nothing else; see
    /// [`TaskSource::set_task_content`].
    ///
    /// One update of the body, which differs from the body GitHub holds only outside the
    /// metadata slot — the slot is kept byte for byte, so every caller key and every list
    /// this source keeps there reads back as it was. A body that would not change is not
    /// sent at all.
    async fn replace_content(
        &self,
        id: &NativeId,
        content: &str,
    ) -> Result<Option<()>, SourceError> {
        let Some(mut item) = self
            .bound_item(id)
            .await?
            .filter(|item| item.kind == BoardKind::Work(ItemKind::Task))
        else {
            return Ok(None);
        };
        let held = item.raw_body.clone().unwrap_or_default();
        // The task's `## Links` section, re-rendered from the links it holds, stays with it.
        let content = &onetaskgraph_plugin_api::relinked(content, item.body.as_deref(), &item.slot);
        let body = with_content(&held, &self.marked(content))?;
        // Checked before anything is sent: content ending in what this source reads as its own
        // metadata slot would read back as metadata rather than as the content it was.
        let (visible, slot) = self.visible(Some(body.clone()))?;
        if visible.as_deref().unwrap_or_default() != content || slot != item.slot {
            return Err(SourceError::Refused {
                message: format!(
                    "this content ends in what source {} reads as its own metadata slot \
                     ({METADATA_OPEN:?}), so part of it would read back as metadata rather than \
                     as content; next: remove that trailing block from the content",
                    self.name
                ),
            });
        }
        if body != held {
            self.update_content(item.content_kind, &item.id, json!({"body": body}))
                .await?;
        }
        item.body = visible.filter(|value| !value.is_empty());
        item.raw_body = Some(body);
        item.slot = slot;
        self.remember_written(item, false)?;
        Ok(Some(()))
    }

    /// Apply one targeted update to one task; see [`TaskSource::update_task`].
    ///
    /// One read of the item — which carries the board's field definitions and the issue's
    /// `blockedBy`, so neither is read again — and then only what differs from it: the
    /// `Status` option and the `Priority` field together in one request, the `blockedBy`
    /// additions and removals the named edges differ by, and last one `updateIssue` carrying
    /// the title, the body — visible content and metadata slot together — and a state change.
    /// So an update naming any of title, body, metadata, status and priority is one read and
    /// at most two writes. The body goes last so that a write refused part-way leaves it, and
    /// the metadata in it, as it stood. A terminal status selects its option and then closes,
    /// as a whole write does; an open one selects its option and then reopens. The origin
    /// field is never written: an update is of an item that already exists, whose origin is
    /// what it is.
    ///
    /// The task answered is the item as those writes left it, built from the read and what was
    /// sent rather than read again — the same record a later read in this run answers from.
    ///
    /// Its `version` is the one the task holds after the writes: the change token the last
    /// mutation's own response spells — each sent as the twin selecting the whole issue, at
    /// the price of the plain document — or, where that mutation shows no whole issue (several
    /// board fields written together, or `blockedBy` alone), the token one read right after
    /// the writes spells. Never the token the item was read at before them. Against GitHub
    /// itself that token has not settled yet (`change_token::SETTLED`), so a read gives none
    /// until it has, and a write held to it before then is refused rather than let through;
    /// once settled, a read gives exactly this token unless the task changed again.
    async fn targeted_update(
        &self,
        id: &NativeId,
        update: &TaskUpdate,
    ) -> Result<Option<TaskUpdateOutcome>, SourceError> {
        Box::pin(SPELLING.scope(
            std::ptr::from_ref(self) as usize,
            self.composed_update(id, update),
        ))
        .await
    }

    /// [`Self::targeted_update`], inside the scope that has its mutations answer whole.
    async fn composed_update(
        &self,
        id: &NativeId,
        update: &TaskUpdate,
    ) -> Result<Option<TaskUpdateOutcome>, SourceError> {
        // Everything this source can refuse without reading the item is refused first, in the
        // words a whole write of the same fields is refused with.
        update.consistent()?;
        if update.assignees.is_some() && !self.assignees {
            return Err(onetaskgraph_plugin_api::unwritable_field(KIND, "assignees"));
        }
        if update
            .title
            .as_deref()
            .is_some_and(|title| title.starts_with(DESIGN_TITLE_PREFIX))
        {
            return Err(SourceError::Refused {
                message: format!(
                    "the title of this task begins {DESIGN_TITLE_PREFIX:?}, which is how source {} \
                     spells a document, so it would read back as one rather than as a task; \
                     retitle it",
                    self.name
                ),
            });
        }
        if let Some(delivers) = &update.delivers {
            TaskRef::listed(
                TaskRef::DELIVERS_KEY,
                id,
                Some(&self.name),
                delivers.clone(),
            )
            .map_err(|message| SourceError::Refused { message })?;
        }
        if self.priorities.is_none()
            && update
                .priority
                .is_some_and(|priority| priority != Priority::None)
        {
            return Err(self.holds_no_priority());
        }
        let target = update
            .status
            .as_ref()
            .map(|status| self.resolved_target(ItemKind::Task, status.category))
            .transpose()?;
        let Some(mut item) = self
            .bound_item(id)
            .await?
            .filter(|item| item.kind == BoardKind::Work(ItemKind::Task))
        else {
            return Ok(None);
        };
        let before = item.task()?;
        // The people it is to be assigned to, by node id, when they are not exactly whom it
        // holds: looked up — unless this process already has — and refused for a login GitHub
        // does not know, before anything is sent.
        let assignee_ids = match &update.assignees {
            Some(people)
                if item
                    .assignees
                    .as_deref()
                    .is_none_or(|held| !assignees::same_people(held, people)) =>
            {
                if item.content_kind == ContentKind::DraftIssue {
                    return Err(self.assigns_a_draft());
                }
                self.resolve_people(people).await?;
                Some(self.people_ids(people, &item.written_name())?)
            }
            _ => None,
        };

        let mut status_move = None;
        if let (Some(status), Some(target)) = (&update.status, target) {
            let board = self.status_board(&item).await?;
            let (field, option, name) = self
                .column_for(&board.fields, ItemKind::Task, status.category, &target)?
                .ok_or_else(|| SourceError::Malformed {
                    message: format!(
                        "status {} of source {} names no board Status option",
                        category_name(status.category),
                        self.name
                    ),
                })?;
            let terminal = matches!(target, StatusTarget::Terminal(_, _));
            if terminal && item.content_kind == ContentKind::DraftIssue {
                return Err(self.closes_a_draft(status.category));
            }
            let landed = match &target {
                StatusTarget::Terminal(_, reason) => {
                    self.statuses
                        .status(ItemKind::Task, Some(&name), true, Some(reason.reason()))
                }
                _ => self
                    .statuses
                    .status(ItemKind::Task, Some(&name), false, None),
            };
            let option_moves = item
                .option
                .as_deref()
                .is_none_or(|held| !held.eq_ignore_ascii_case(&name));
            let state_moves = item.content_kind == ContentKind::Issue
                && (item.closed != terminal || (terminal && item.status != landed));
            if let Some(moves) = Moves::of(option_moves, state_moves) {
                status_move = Some(StatusMove {
                    board: board.id,
                    field,
                    option,
                    name,
                    target,
                    landed,
                    moves,
                });
            }
        }

        let mut priority_move = None;
        if let Some(priority) = update.priority
            && self.priorities.is_some()
            && item.priority != HeldPriority::Read(priority)
        {
            let board = match (item.carried_board(), item.named_board()) {
                (Some(board), _) => board,
                (None, Some(board)) if item.defines(PRIORITY_FIELD) => BoardFields {
                    id: board,
                    fields: json!({"nodes": item.fields, "pageInfo": {"hasNextPage": false}}),
                },
                _ => self.board_fields().await?,
            };
            if let Some(write) = self.priority_write(&board.fields, Some(&item), priority)? {
                priority_move = Some((board.id, write, priority));
            }
        }

        // The metadata keys this update names that a board text field projects, and what
        // that does to those fields — refused, as a whole write refuses it, before anything
        // is sent.
        let touched: Vec<&str> = update
            .metadata_set
            .keys()
            .chain(update.metadata_remove.iter())
            .map(MetadataKey::as_str)
            .collect();
        let mut projection_move = None;
        if self
            .metadata_fields
            .iter()
            .any(|projection| touched.contains(&&*projection.key))
        {
            let mut metadata = item.slot.clone();
            for (key, value) in &update.metadata_set {
                metadata.insert(key.as_str().to_owned(), value.clone());
            }
            for key in &update.metadata_remove {
                metadata.remove(key.as_str());
            }
            let board = self.projection_board(&item).await?;
            let (writes, projected) = self.projection_writes(
                &board.fields,
                Some(&item.projected),
                &metadata,
                Some(&touched),
            )?;
            projection_move = Some((board.id, writes, projected));
        }

        // Resolved before the body is composed, because a far end `blockedBy` cannot name is
        // recorded in the slot, and the slot travels in the one body update below.
        let edges = match &update.depends_on {
            Some(edges) => Some(
                self.partition_edges(
                    BoardKind::Work(ItemKind::Task),
                    item.content_kind,
                    item.blocked_by.as_deref(),
                    edges,
                )
                .await?,
            ),
            None => None,
        };

        // A record primed from what a producer wrote holds no body anybody read, so an update
        // changing the body reads it first: by the `Status` write's own answer when the update
        // moves the option — the mutation answering with the issue whole — and otherwise by
        // reading the issue. The body is then composed against what GitHub holds now.
        let mut option_selected = false;
        // The mark stays until a read lands, so an update that writes no body, or one whose
        // read fails, leaves the next update of it to read the body too.
        if (update.content.is_some()
            || !update.metadata_set.is_empty()
            || !update.metadata_remove.is_empty()
            || update.delivers.is_some()
            || update.repositories.is_some()
            || edges.is_some())
            && self.is_unread(id)
        {
            let live = match status_move.as_ref().filter(|moving| moving.moves.option()) {
                Some(moving) => {
                    option_selected = true;
                    Some(
                        self.select_reading_body(
                            moving.board.as_str(),
                            &item.item_id,
                            &moving.field,
                            &moving.option,
                        )
                        .await?,
                    )
                }
                None => {
                    self.forget(id)?;
                    let Some(read) = self.item_by_id(id).await? else {
                        return Ok(None);
                    };
                    read.raw_body
                }
            };
            let (visible, read_slot) = self.visible(live.clone())?;
            self.mark_read(id);
            item.raw_body = live;
            item.body = visible;
            item.slot = read_slot;
        }

        let mut slot = item.slot.clone();
        for (key, value) in &update.metadata_set {
            slot.insert(key.as_str().to_owned(), value.clone());
        }
        for key in &update.metadata_remove {
            slot.remove(key.as_str());
        }
        if let Some(delivers) = &update.delivers {
            set_task_list(&mut slot, TaskRef::DELIVERS_KEY, delivers);
        }
        // The list a whole write records, by the same rule: none when it is exactly the issue's
        // own repository, or empty with no issue repository to derive, and otherwise the list.
        if let Some(repositories) = &update.repositories {
            let derivable = item
                .own_repository
                .as_ref()
                .map_or(repositories.is_empty(), |own| {
                    repositories.as_slice() == std::slice::from_ref(own)
                });
            if derivable {
                slot.remove(Repository::METADATA_KEY);
            } else {
                slot.insert(
                    Repository::METADATA_KEY.to_owned(),
                    Value::Array(
                        repositories
                            .iter()
                            .map(|repository| Value::String(repository.as_str().to_owned()))
                            .collect(),
                    ),
                );
            }
        }
        if let Some((_, recorded)) = &edges {
            record_edges(&mut slot, recorded);
        }
        let held = item.raw_body.clone().unwrap_or_default();
        // A content write keeps the task's `## Links` section, re-rendered from the links the
        // slot holds once written.
        let relinked = update
            .content
            .as_deref()
            .map(|content| onetaskgraph_plugin_api::relinked(content, item.body.as_deref(), &slot));
        let content = match &relinked {
            Some(content) => with_content(&held, &self.marked(content))?,
            None => held.clone(),
        };
        // A slot holding what it held is kept byte for byte, compared as JSON rather than as
        // the body's bytes, as a metadata write compares it: a slot a person spelled with
        // other whitespace would otherwise be re-encoded, which is a write nobody asked for.
        let body = if slot == item.slot {
            content
        } else {
            with_slot(&content, &slot)?
        };
        // Checked before anything is sent, as a content write checks it: content ending in
        // what this source reads as its own slot would read back as metadata.
        let (visible, read) = self.visible(Some(body.clone()))?;
        let wanted = relinked.as_deref().or(item.body.as_deref());
        if visible.as_deref().unwrap_or_default() != wanted.unwrap_or_default() || read != slot {
            return Err(SourceError::Refused {
                message: format!(
                    "this content ends in what source {} reads as its own metadata slot \
                     ({METADATA_OPEN:?}), so part of it would read back as metadata rather than \
                     as content; next: remove that trailing block from the content",
                    self.name
                ),
            });
        }
        let recorded_moves =
            slot.get(DependencyEdge::RECORDED_KEY) != item.slot.get(DependencyEdge::RECORDED_KEY);

        // One `updateIssue` carries all three, because every mutation spends the secondary
        // limiter and the title, body and state are one mutation's inputs.
        let mut fields = serde_json::Map::new();
        if let Some(title) = update.title.as_ref().filter(|title| **title != item.title) {
            fields.insert("title".to_owned(), json!(title));
        }
        if body != held {
            fields.insert("body".to_owned(), json!(body));
        }
        if let Some(moving) = status_move.as_ref().filter(|moving| moving.moves.state()) {
            fields.insert("stateInput".to_owned(), state_input(Some(&moving.target)));
        }
        // The assignees ride the same mutation: replacing whoever the issue holds, in the one
        // `updateIssue` this update already ends with — or as the one it sends for them alone.
        if let Some(ids) = &assignee_ids {
            fields.insert("assigneeIds".to_owned(), json!(ids));
        }
        // **The body is written last, and that is the guarantee a refusal part-way keeps.**
        // GitHub runs no two requests as one, and runs one document's mutation fields in order
        // without undoing an earlier field when a later one fails — so a body written before a
        // board field the board then refused would be left changed. Written after every other
        // write has landed, a refusal anywhere leaves the item's body, and every metadata key
        // it carries, exactly as they stood. So the `Status` option and the `Priority` field go
        // first, together in one request — a terminal option selected before the issue
        // closes, as a whole write does — then the `blockedBy` difference, then the body.
        let mut board_writes: Vec<(&BoardId, (String, Value))> = Vec::new();
        let mut clears: Vec<(&BoardId, &str)> = Vec::new();
        if let Some(moving) = status_move
            .as_ref()
            .filter(|moving| moving.moves.option() && !option_selected)
        {
            board_writes.push((
                &moving.board,
                (
                    moving.field.clone(),
                    json!({"singleSelectOptionId": moving.option}),
                ),
            ));
        }
        match &priority_move {
            Some((board, PriorityWrite::Select { field, option }, _)) => board_writes.push((
                board,
                (field.clone(), json!({"singleSelectOptionId": option})),
            )),
            Some((board, PriorityWrite::Clear { field }, _)) => clears.push((board, field)),
            None => {}
        }
        if let Some((board, writes, _)) = &projection_move {
            for write in writes {
                let (mut values, mut cleared) = (Vec::new(), Vec::new());
                write.join(&mut values, &mut cleared);
                board_writes.extend(values.into_iter().map(|value| (board, value)));
                clears.extend(cleared.into_iter().map(|field| (board, field)));
            }
        }
        let mut boards: Vec<&BoardId> = board_writes.iter().map(|(board, _)| *board).collect();
        boards.extend(clears.iter().map(|(board, _)| *board));
        boards.dedup_by(|one, other| one.as_str() == other.as_str());
        for board in boards {
            let writes = board_writes
                .iter()
                .filter(|(on, _)| on.as_str() == board.as_str())
                .map(|(_, write)| write.clone())
                .collect::<Vec<_>>();
            let cleared = clears
                .iter()
                .filter(|(on, _)| on.as_str() == board.as_str())
                .map(|(_, field)| *field)
                .collect::<Vec<_>>();
            self.set_item_fields(board.as_str(), &item.item_id, &writes, &cleared)
                .await?;
        }
        // Whether anything is sent: what the version answered has to come after.
        let sends =
            option_selected || !board_writes.is_empty() || !clears.is_empty() || !fields.is_empty();
        let mut blocked_by_moved = false;
        if let Some((native, _)) = &edges
            && item.content_kind == ContentKind::Issue
        {
            blocked_by_moved = self
                .reconcile_blocked_by(
                    &item.id,
                    native,
                    Issue::Existing(item.blocked_by.as_deref()),
                )
                .await?;
        }
        if !fields.is_empty() {
            self.update_content(item.content_kind, &item.id, Value::Object(fields))
                .await?;
        }

        // The `blockedBy` this item was read with is what the write just changed, so it is no
        // answer any more: a later write in this command would reconcile against it — re-adding
        // what it kept and never removing what it dropped — and a forward-edge read would
        // answer with it. Left unknown, each asks GitHub instead.
        if blocked_by_moved {
            item.blocked_by = None;
        }
        if let Some(title) = &update.title {
            item.title.clone_from(title);
        }
        item.body = visible.filter(|value| !value.is_empty());
        item.raw_body = (!body.is_empty() || item.raw_body.is_some()).then_some(body);
        item.slot = slot;
        if let Some(delivers) = &update.delivers {
            item.delivers.clone_from(delivers);
        }
        if let Some(repositories) = &update.repositories {
            item.repositories.clone_from(repositories);
        }
        if let Some(moving) = status_move {
            item.closed = matches!(moving.target, StatusTarget::Terminal(_, _))
                && item.content_kind == ContentKind::Issue;
            item.status = moving.landed;
            item.option = Some(moving.name);
        }
        if let Some((_, _, priority)) = priority_move {
            item.priority = HeldPriority::Read(priority);
        }
        if let Some((_, _, projected)) = projection_move {
            item.projected = projected;
        }
        // Only a write that assigned anyone moves whom the item holds, and they are who GitHub
        // says they are; one naming the people it already holds leaves its read as it was.
        if let (Some(people), Some(_)) = (&update.assignees, &assignee_ids) {
            item.assignees = Some(self.spelled(people));
        }
        let mut task = item.task()?;
        let mut written = update.changed(&before, &task);
        if blocked_by_moved || recorded_moves {
            written.insert(UpdatedField::DependsOn);
        }
        if sends || blocked_by_moved {
            // The token the item was read at is not the one it holds now. A batch of updates
            // holding this one's writes back reads every task it wrote once they are sent.
            task.version = if self.holding_writes_of(id) {
                None
            } else {
                self.version_after_write(id).await
            };
        }
        self.remember_written(item, false)?;
        Ok(Some(TaskUpdateOutcome {
            task,
            written,
            delivers_before: before.delivers,
        }))
    }

    /// The change token the task `id` holds after the writes a targeted update just sent: the
    /// one the last mutation's own response showed, or, where none showed the whole issue, the
    /// one a read made now spells. `None` when that read cannot say — never the token the
    /// task was read at before the writes.
    async fn version_after_write(&self, id: &NativeId) -> Option<String> {
        // A response that showed the issue whole answers, a token or none: a read made now
        // would read the same parts.
        if let Some(answered) = self.written_answer(id) {
            return answered;
        }
        // An item this command created is answered from its own record, not read.
        if self.created().ok()?.iter().any(|own| own.id == *id) {
            return None;
        }
        let read = self.item_by_id(id).await.ok()??;
        let token = read.written_version()?;
        self.remember_written_token(&read, token.clone());
        Some(token)
    }

    /// Replace one issue's visible body and its [`MetadataKey::TEMPLATE_KEY`] slot entry
    /// together, and nothing else; see [`TaskSource::set_task_rendering`].
    ///
    /// One update of the body: the content outside the slot, and inside it that one entry,
    /// every other entry kept as it was. This source keeps no template answers — an issue has
    /// no room beside itself that is not its body, and answers written there would duplicate
    /// what the content already says and count against GitHub's body limit — so `answers`
    /// reaches nothing here. A body that would not change is not sent at all.
    async fn replace_rendering(
        &self,
        id: &NativeId,
        kind: BoardKind,
        content: &str,
        provenance: &Value,
        assets: Option<&onetaskgraph_plugin_api::AssetWrite>,
    ) -> Result<Option<onetaskgraph_plugin_api::AssetsWritten>, SourceError> {
        let Some(mut item) = self.bound_item(id).await?.filter(|item| item.kind == kind) else {
            return Ok(None);
        };
        let held = item.raw_body.clone().unwrap_or_default();
        let mut slot = item.slot.clone();
        slot.insert(MetadataKey::TEMPLATE_KEY.to_owned(), provenance.clone());
        let rewritten;
        let content = if let Some(assets) = assets {
            let uploads = self
                .upload_assets(item.own_repository.as_ref(), assets)
                .await?;
            rewritten =
                onetaskgraph_plugin_api::serve_asset_references(content, &mut slot, &uploads);
            rewritten.as_str()
        } else {
            content
        };
        let relinked = onetaskgraph_plugin_api::relinked(content, item.body.as_deref(), &slot);
        let content = relinked.as_str();
        let body = with_slot(&with_content(&held, &self.marked(content))?, &slot)?;
        // Checked before anything is sent, as a content write checks it.
        let (visible, read) = self.visible(Some(body.clone()))?;
        if visible.as_deref().unwrap_or_default() != content || read != slot {
            return Err(SourceError::Refused {
                message: format!(
                    "this content ends in what source {} reads as its own metadata slot \
                     ({METADATA_OPEN:?}), so part of it would read back as metadata rather than \
                     as content; next: remove that trailing block from the template",
                    self.name
                ),
            });
        }
        if body != held {
            self.update_content(item.content_kind, &item.id, json!({"body": body}))
                .await?;
        }
        item.body = visible.filter(|value| !value.is_empty());
        item.raw_body = Some(body);
        // The content as a read answers it: without the `## Links` section of a linked task.
        let answered = onetaskgraph_plugin_api::read_content(Some(content.to_owned()), &read);
        item.slot = read;
        self.remember_written(item, false)?;
        Ok(Some(onetaskgraph_plugin_api::AssetsWritten {
            id: id.clone(),
            content: answered,
        }))
    }

    async fn set_item_field(
        &self,
        board_id: &str,
        item_id: &str,
        field_id: &str,
        value: Value,
    ) -> Result<(), SourceError> {
        let data = self
            .graphql(
                graphql::UPDATE_FIELD,
                json!({"input":{
                    "projectId":board_id,"itemId":item_id,"fieldId":field_id,"value":value
                },"readPriority":false,"priorityName":PRIORITY_FIELD}),
            )
            .await?;
        field_update_item(&data, item_id).map(|_| ())
    }

    /// GitHub accepts one value per field mutation; aliases combine those mutations in
    /// one request — writes in the order given, then clears. A lone write is
    /// [`graphql::UPDATE_FIELD`] and a lone clear [`graphql::CLEAR_FIELD`]; more than one
    /// request's slots hold go in further requests of the same document, in order. Every
    /// returned item id is checked, including optional aliases.
    async fn set_item_fields(
        &self,
        board: &str,
        item: &str,
        fields: &[(String, Value)],
        clears: &[&str],
    ) -> Result<(), SourceError> {
        if !(fields.is_empty() && clears.is_empty())
            && self.hold_fields(board, item, fields, clears)?
        {
            return Ok(());
        }
        match (fields, clears) {
            ([], []) => return Ok(()),
            ([(field, value)], []) => {
                return self.set_item_field(board, item, field, value.clone()).await;
            }
            ([], [field]) => {
                return self
                    .write_priority(
                        board,
                        item,
                        &PriorityWrite::Clear {
                            field: (*field).to_owned(),
                        },
                    )
                    .await;
            }
            _ => {}
        }
        // Any field of the request stands in for an input a slot leaves out, which GitHub
        // still requires to be well formed although it never runs it.
        let filler = fields
            .first()
            .map_or_else(|| clears[0], |(field, _)| field.as_str());
        let mut writes = fields.iter();
        let mut cleared = clears.iter();
        loop {
            let batch: Vec<&(String, Value)> =
                writes.by_ref().take(FIELD_WRITE_SLOTS.len()).collect();
            let batch_clears: Vec<&&str> = cleared.by_ref().take(FIELD_CLEAR_SLOTS.len()).collect();
            if batch.is_empty() && batch_clears.is_empty() {
                return Ok(());
            }
            let mut variables = serde_json::Map::new();
            for (index, slot) in FIELD_WRITE_SLOTS.iter().enumerate() {
                let input = match batch.get(index) {
                    Some((field, value)) => {
                        json!({"projectId":board,"itemId":item,"fieldId":field,"value":value})
                    }
                    None => json!({"projectId":board,"itemId":item,"fieldId":filler,
                                   "value":{"text":""}}),
                };
                variables.insert(slot.variable.to_owned(), input);
                variables.insert(slot.include.to_owned(), json!(index < batch.len()));
            }
            for (index, slot) in FIELD_CLEAR_SLOTS.iter().enumerate() {
                let field = batch_clears.get(index).map_or(filler, |field| **field);
                variables.insert(
                    slot.variable.to_owned(),
                    json!({"projectId":board,"itemId":item,"fieldId":field}),
                );
                variables.insert(slot.include.to_owned(), json!(index < batch_clears.len()));
            }
            let data = self
                .graphql(graphql::UPDATE_FIELDS, Value::Object(variables))
                .await?;
            let sent = FIELD_WRITE_SLOTS[..batch.len()]
                .iter()
                .chain(&FIELD_CLEAR_SLOTS[..batch_clears.len()]);
            for slot in sent {
                let alias = slot.alias;
                let returned = data
                    .get(alias)
                    .and_then(|value| value.get("projectV2Item"))
                    .ok_or_else(|| SourceError::Malformed {
                        message: format!("GitHub field update {alias} returned no project item"),
                    })?;
                if required_str(returned, "id")? != item {
                    return Err(SourceError::Malformed {
                        message: format!(
                            "GitHub field update {alias} returned the wrong project item"
                        ),
                    });
                }
            }
        }
    }

    async fn native_dependency_ids(&self, id: &NativeId) -> Result<Vec<String>, SourceError> {
        let mut after: Option<String> = None;
        let mut ids = Vec::new();
        loop {
            let data = self
                .graphql(
                    graphql::ISSUE_DEPENDENCIES,
                    json!({"id":id.0,"first":MAX_PAGE_SIZE,"after":after}),
                )
                .await?;
            let connection =
                data.pointer("/node/blockedBy")
                    .ok_or_else(|| SourceError::Malformed {
                        message: "GitHub dependency response has no blockedBy connection".into(),
                    })?;
            ids.extend(
                connection
                    .get("nodes")
                    .and_then(Value::as_array)
                    .ok_or_else(|| SourceError::Malformed {
                        message: "GitHub dependency response nodes is not an array".into(),
                    })?
                    .iter()
                    .map(|value| required_str(value, "id").map(str::to_owned))
                    .collect::<Result<Vec<_>, _>>()?,
            );
            let next = next_cursor(connection)?;
            if let Some(next) = &next {
                validate_cursor_progress(after.as_deref(), &next.0)?;
            }
            after = next.map(|cursor| cursor.0);
            if after.is_none() {
                return Ok(ids);
            }
        }
    }

    async fn dependencies(
        &self,
        id: &NativeId,
        near_kind: ItemKind,
        direction: Direction,
        page: &PageRequest,
    ) -> Result<Page<DependencyEdge>, SourceError> {
        validate_page(page)?;
        let limit = page.limit.min(MAX_PAGE_SIZE) as usize;
        let cursor = page.cursor.as_ref().map(|c| c.0.as_str());
        let recorded = recorded_offset(cursor, direction)?;
        // What this issue is blocked by, when a read of it by its own id in this command
        // already carried the whole connection — a copy reads the item it writes before it
        // reads its edges — and the page asked for is the whole of it, or the recorded tail
        // after it. Answered from that read, in the shape the dependency read answers in;
        // anything else is asked of GitHub.
        let carried = match direction {
            Direction::DependsOn => self
                .resolved_cache()?
                .get(id)
                .filter(|item| item.content_kind == ContentKind::Issue)
                .and_then(|item| Some((item.blocked_by.clone()?, item.raw_body.clone()))),
            Direction::DependedOnBy => None,
        }
        .filter(|(nodes, _)| recorded.is_some() || (cursor.is_none() && nodes.len() <= limit));
        // Asked for even in the recorded phase, whose page reads nothing from the
        // connection: `__typename` is what says whether this item has a native
        // relationship at all, and that is what decides which far ends the reserved key is
        // allowed to hold.
        let data = match carried {
            Some((nodes, body)) => json!({"node":{"__typename":"Issue","body":body,
                "blockedBy":{"nodes":nodes,"pageInfo":{"hasNextPage":false,"endCursor":null}}}}),
            None => {
                let after = if recorded.is_some() { None } else { cursor };
                let first = page.limit.min(MAX_PAGE_SIZE);
                match direction {
                    Direction::DependsOn => {
                        self.graphql(
                            graphql::ISSUE_DEPENDS_ON,
                            json!({"id":id.0,"first":first,"after":after,"owner":self.owner,
                                   "number":self.project_number,
                                   "boardItems":BOARD_ITEMS_PAGE_SIZE}),
                        )
                        .await?
                    }
                    Direction::DependedOnBy => {
                        self.graphql(
                            graphql::ISSUE_DEPENDED_ON_BY,
                            json!({"id":id.0,"first":first,"after":after}),
                        )
                        .await?
                    }
                }
            }
        };
        let draft_slot = draft_slot(&data, id, self.project_number)?;
        let node =
            data.get("node")
                .filter(|v| !v.is_null())
                .ok_or_else(|| SourceError::Refused {
                    message: format!(
                        "GitHub item {} was not found or does not support dependencies",
                        id.0
                    ),
                })?;
        let connection_name = match direction {
            Direction::DependsOn => "blockedBy",
            Direction::DependedOnBy => "blocking",
        };
        // A draft has neither `blockedBy` nor `blocking`, so nothing it depends on can be
        // named natively and the reserved key may hold any far end. An issue's connections
        // hold issues, and this source reads them at the near item's own level.
        let natively_names = (required_str(node, "__typename")? == "Issue").then_some(near_kind);
        if let Some(offset) = recorded {
            return Ok(recorded_page(
                self.recorded_edges(
                    id,
                    near_kind,
                    direction,
                    natively_names,
                    node,
                    draft_slot.as_ref(),
                )
                .await?,
                offset,
                limit,
            ));
        }
        if natively_names.is_none() {
            return Ok(recorded_page(
                self.recorded_edges(
                    id,
                    near_kind,
                    direction,
                    natively_names,
                    node,
                    draft_slot.as_ref(),
                )
                .await?,
                0,
                limit,
            ));
        }
        let connection = node
            .get(connection_name)
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub dependency response is missing its connection".into(),
            })?;
        let nodes = connection
            .get("nodes")
            .and_then(Value::as_array)
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub dependency response nodes is not an array".into(),
            })?;
        // `from` depends on `to`, always. GitHub spells the same relationship from either
        // end — `blockedBy` lists what this item waits on, `blocking` lists what waits on
        // it — so the near item is `from` in one direction and `to` in the other.
        let items = nodes
            .iter()
            .map(|value| {
                let related = NativeId(required_str(value, "id")?.into());
                let related_kind = related_kind(value, self.sub_issues)?;
                let (from, to) = match direction {
                    Direction::DependsOn => (
                        DependencyEndpoint::from_native(id.clone(), near_kind),
                        DependencyEndpoint::from_native(related, related_kind),
                    ),
                    Direction::DependedOnBy => (
                        DependencyEndpoint::from_native(related, related_kind),
                        DependencyEndpoint::from_native(id.clone(), near_kind),
                    ),
                };
                Ok(DependencyEdge {
                    from,
                    to,
                    kind: DependencyKind::Blocks,
                })
            })
            .collect::<Result<Vec<_>, SourceError>>()?;
        let mut next = next_cursor(connection)?;
        if let Some(next) = &next {
            validate_cursor_progress(cursor, &next.0)?;
        }
        // The recorded far ends ride on the last page of `blockedBy`, whose answer carried the
        // body they are recorded in, as far as the page has room — so they cost no request of
        // their own — and only what does not fit is resumed by a cursor.
        let mut items = items;
        if next.is_none() {
            let recorded = self
                .recorded_edges(
                    id,
                    near_kind,
                    direction,
                    natively_names,
                    node,
                    draft_slot.as_ref(),
                )
                .await?;
            let room = limit.saturating_sub(items.len());
            let total = recorded.len();
            items.extend(recorded.into_iter().take(room));
            if total > room {
                next = Some(Cursor(format!("{RECORDED_CURSOR}{room}")));
            }
        }
        Ok(Page { items, next })
    }

    /// The edges this item records under [`DependencyEdge::RECORDED_KEY`], which is where
    /// a far end in another source has to live: no GitHub issue relationship can name one.
    ///
    /// Only forwards. The reverse of a recorded edge is derived from the far end, and this
    /// source never writes one down.
    ///
    /// The metadata lives in the item's own body slot, and `node` is the dependency read's
    /// own answer, which carries an issue's body — so an issue's recorded edges cost no
    /// request beyond the read already made, and reading the board for them would be a
    /// walk of every item for one field of one. A draft has no body in that answer, because
    /// a draft is not an issue, so a draft's are read off its own read by id — never off a
    /// listing of the board, which can be behind on the very item asked about.
    async fn recorded_edges(
        &self,
        id: &NativeId,
        near_kind: ItemKind,
        direction: Direction,
        natively_names: Option<ItemKind>,
        node: &Value,
        draft_slot: Option<&Option<BTreeMap<String, Value>>>,
    ) -> Result<Vec<DependencyEdge>, SourceError> {
        if direction != Direction::DependsOn {
            return Ok(Vec::new());
        }
        let slot = match (node.get("body"), draft_slot) {
            (Some(body), _) if natively_names.is_some() => {
                metadata_body(body.as_str().map(str::to_owned))?.1
            }
            // A draft the dependency read itself answered: this board's, or none of its.
            (_, Some(Some(slot))) => slot.clone(),
            (_, Some(None)) => return Ok(Vec::new()),
            _ => {
                let Some(item) = self.bound_item(id).await? else {
                    return Ok(Vec::new());
                };
                item.slot
            }
        };
        DependencyEdge::recorded(&slot, id, near_kind, &self.name, natively_names)
            .map_err(|message| SourceError::Malformed { message })
    }

    /// Why `creation_owners` refuses a create filed in `target`, or `None` when it allows it —
    /// every create when the list is empty.
    fn owner_refusal(&self, target: &RepositoryTarget) -> Option<String> {
        if self.creation_owners.is_empty()
            || self
                .creation_owners
                .iter()
                .any(|owner| owner.eq_ignore_ascii_case(&target.owner))
        {
            return None;
        }
        Some(format!(
            "source {} would create the issue in repository {}, owned by {}, and its \
             creation_owners allows only repositories of {}",
            self.name,
            target.slug(),
            target.owner,
            self.creation_owners.join(", ")
        ))
    }

    fn configured_repository(&self) -> Result<&RepositoryTarget, SourceError> {
        self.repository
            .as_ref()
            .ok_or_else(|| SourceError::Refused {
                message: format!(
                    "source {} has no repository configured, and a GitHub Projects board has no \
                 repository of its own to create an issue in; set repository: owner/name on \
                 this source",
                    self.name
                ),
            })
    }

    /// The repository one new issue is created in, under the rule [`RepositoryTarget`]
    /// states.
    ///
    /// The fallback is demanded first, whichever arm answers: a write without a configured
    /// repository is refused naming the field exactly as it was before the rule existed,
    /// so a source that could not write before cannot write now, rather than writing for
    /// the one item whose own field happens to decide it.
    ///
    /// Everything this refuses is refused before `createIssue`, so a refusal leaves no
    /// issue behind: an entry that is not a repository on [`RepositoryTarget::HOST`], an
    /// entry owned by someone other than the owner of the parent issue's repository —
    /// GitHub accepts a sub-issue from another repository of the same owner and from no
    /// other, so `addSubIssue` would refuse it after the issue existed — a parent the
    /// board does not hold, and a parent that is a draft, which GitHub gives no sub-issues,
    /// both of which `addSubIssue` would likewise refuse too late. Whether the entry exists
    /// and is visible to the token is checked where its node id is resolved, still before
    /// `createIssue`. The parent is read by its own id through [`Self::item_by_id`] — never
    /// looked up in a listing of the board, which can be minutes behind an issue its own
    /// `projectItems` already places on it — and that read answers first from this process's
    /// own record, so a project created moments ago in this command answers though GitHub
    /// has not caught up.
    async fn creation_target(
        &self,
        incoming: &Incoming<'_>,
    ) -> Result<RepositoryTarget, SourceError> {
        let what = format!(
            "{} {:?}",
            incoming.written.kind().describes(),
            incoming.title
        );
        let target = self
            .creation_rule(incoming.repositories, incoming.parent, &what)
            .await?;
        // The engine asks `creation_refusal` before anything of a copy is written; this holds
        // every other create to the same restriction, before `createIssue`.
        match self.owner_refusal(&target) {
            Some(message) => Err(SourceError::Refused { message }),
            None => Ok(target),
        }
    }

    /// [`Self::creation_target`]'s rule over what it reads of the item — its `repositories`
    /// and the project issue it is filed under — with `what` naming the item in a refusal: the
    /// one place the rule is, so the create, [`TaskSource::creation_repository`] and the
    /// write-time read of who can read the issue ([`TaskSource::visibility`]) cannot answer
    /// differently.
    ///
    /// On a source declared private, an item's one named repository is read for its
    /// visibility first — once per engine hold, see the `visibility` module — and one that
    /// does not read private decides nothing: the issue is created in the parent's
    /// repository, else the configured one. A read that cannot be answered refuses.
    async fn creation_rule(
        &self,
        repositories: &[Repository],
        parent: Option<&NativeId>,
        what: &str,
    ) -> Result<RepositoryTarget, SourceError> {
        let parent = match parent {
            Some(parent) => Some(self.bound_item(parent).await?.ok_or_else(|| {
                SourceError::Refused {
                    message: format!(
                        "GitHub project issue {} was not found on the board of source {}, so {} \
                         cannot be filed under it",
                        parent.0, self.name, what
                    ),
                }
            })?),
            None => None,
        };
        let parents_repository = parent
            .as_ref()
            .map(|parent| {
                // A draft is on the board and so is found, but it has no repository to
                // place a task in and GitHub gives it no sub-issues, so `addSubIssue`
                // would refuse the task only once `createIssue` had made it.
                if parent.content_kind == ContentKind::DraftIssue {
                    return Err(SourceError::Refused {
                        message: format!(
                            "GitHub project item {} on the board of source {} is a draft, \
                             which cannot have sub-issues, so {} cannot be filed under it",
                            parent.id.0, self.name, what
                        ),
                    });
                }
                // An issue's repository is where a sub-issue is placed and whose owner it
                // is compared against, so a parent whose repository this source cannot
                // spell as `owner/name` — GitHub's login grammar is wider than this
                // source's floor — is one nothing can be filed under.
                parent
                    .own_repository
                    .as_ref()
                    .and_then(|origin| RepositoryTarget::from_origin(origin).ok())
                    .ok_or_else(|| SourceError::Malformed {
                        message: format!(
                            "GitHub project issue {} on the board of source {} is in {}, which \
                             is not a {}/owner/name repository this source can place {} in",
                            parent.id.0,
                            self.name,
                            parent
                                .own_repository
                                .as_ref()
                                .map_or("no repository", Repository::as_str),
                            RepositoryTarget::HOST,
                            what
                        ),
                    })
            })
            .transpose()?;
        self.creation_rule_in_repository(repositories, parents_repository, what)
            .await
    }

    async fn creation_rule_in_repository(
        &self,
        repositories: &[Repository],
        parents_repository: Option<RepositoryTarget>,
        what: &str,
    ) -> Result<RepositoryTarget, SourceError> {
        let fallback = self.configured_repository()?;
        match repositories {
            [named] => {
                let target =
                    RepositoryTarget::from_origin(named).map_err(|_| SourceError::Refused {
                        message: format!(
                            "{} names repository {}, which is not a {}/owner/name repository \
                             source {} can create an issue in; name one that is, or name none",
                            what,
                            named.as_str(),
                            RepositoryTarget::HOST,
                            self.name
                        ),
                    })?;
                // On a source declared private, an issue whose one named repository does not
                // read private is created where an item naming none is: the item still names
                // that repository in its metadata slot, and its issue stays private.
                if self.is_declared_private()
                    && self.repository_visibility(&target).await?
                        != onetaskgraph_plugin_api::Visibility::Private
                {
                    return Ok(parents_repository.unwrap_or_else(|| fallback.clone()));
                }
                if let Some(parents) = &parents_repository
                    && parents.owner != target.owner
                {
                    return Err(SourceError::Refused {
                        message: format!(
                            "{} names repository {}, owned by {}, but its project's issue is in \
                             {}, owned by {}, and GitHub files a sub-issue only in a repository \
                             of the same owner as its parent issue; name a repository of {}, or \
                             name none",
                            what,
                            target.slug(),
                            target.owner,
                            parents.slug(),
                            parents.owner,
                            parents.owner
                        ),
                    });
                }
                Ok(target)
            }
            _ => Ok(parents_repository.unwrap_or_else(|| fallback.clone())),
        }
    }

    /// The node id of the repository `incoming` is being created in, or the refusal naming
    /// the item and the repository the token cannot see.
    ///
    /// Resolved once per command per repository; see [`Self::repository_cache`].
    async fn repository_id(
        &self,
        repository: &RepositoryTarget,
        incoming: &Incoming<'_>,
    ) -> Result<String, SourceError> {
        if let Some(id) = self.repository_cache()?.get(repository).cloned() {
            return Ok(id);
        }
        let data = self
            .graphql(
                graphql::REPOSITORY,
                json!({"owner":repository.owner,"name":repository.name}),
            )
            .await?;
        self.repository_read(&data, repository, incoming)
    }

    /// The repository's node id out of an answer carrying the `repository` root, held for
    /// the rest of this command, or the refusal naming the item that cannot be created in it.
    fn repository_read(
        &self,
        data: &Value,
        repository: &RepositoryTarget,
        incoming: &Incoming<'_>,
    ) -> Result<String, SourceError> {
        let node = data
            .get("repository")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Refused {
                message: format!(
                    "GitHub repository {} was not found or is not visible to the token, so {} \
                     {:?} cannot be created in it",
                    repository.slug(),
                    incoming.written.kind().describes(),
                    incoming.title
                ),
            })?;
        let id = required_str(node, "id")?.to_owned();
        self.repository_cache()?
            .insert(repository.clone(), id.clone());
        Ok(id)
    }

    fn repository_cache(
        &self,
    ) -> Result<std::sync::MutexGuard<'_, BTreeMap<RepositoryTarget, String>>, SourceError> {
        self.repository_cache
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "this source's record of the destination repository was left \
                          inconsistent by an earlier failure; next: run the command again"
                    .into(),
            })
    }

    /// Create or update one board item, whichever kind it is.
    async fn write_item(
        &self,
        incoming: &Incoming<'_>,
        target: Option<&NativeId>,
        depends_on: &[DependencyEdge],
    ) -> Result<onetaskgraph_plugin_api::AssetsWritten, SourceError> {
        let prepared = self.prepare_write(incoming, target, depends_on).await?;
        let landed = match &prepared.destination {
            // Its content is written last, by `complete_write`, once everything else has landed.
            WriteDestination::Existing(item) => Landed {
                content_id: item.id.clone(),
                item_id: Some(item.item_id.clone()),
                url: item.url.clone(),
                number: item.number,
            },
            WriteDestination::New(target) => {
                if self.deferring()? {
                    self.create_filed_issue(
                        prepared.board.id.as_str(),
                        target,
                        incoming,
                        &prepared.body,
                    )
                    .await?
                } else {
                    self.create_and_file_issue(
                        prepared.board.id.as_str(),
                        target,
                        incoming,
                        &prepared.body,
                    )
                    .await?
                }
            }
        };
        self.complete_write(incoming, prepared, landed).await
    }

    /// Everything one write of `incoming` reads and decides before it sends anything: what is
    /// there now, where a new item is created, the board's fields, and every value refused while
    /// nothing has been written — the item's body included.
    async fn prepare_write(
        &self,
        incoming: &Incoming<'_>,
        target: Option<&NativeId>,
        depends_on: &[DependencyEdge],
    ) -> Result<Prepared, SourceError> {
        // Refused before anything is read or written: a task or a project titled the way
        // this board spells a document would land as an issue this same source reads back
        // as a document, so the field this destination cannot carry is named rather than
        // written and silently reclassified.
        if let Written::Work(kind, _) = incoming.written
            && incoming.title.starts_with(DESIGN_TITLE_PREFIX)
        {
            return Err(SourceError::Refused {
                message: format!(
                    "the title of this {} begins {DESIGN_TITLE_PREFIX:?}, which is how source {} \
                     spells a document, so it would read back as one rather than as a {}; \
                     retitle it, or copy it as a document",
                    kind.marker(),
                    self.name,
                    kind.marker()
                ),
            });
        }
        // The destination is read by its own id, and whether this board holds it is decided
        // by that read — its own `projectItems` — rather than by whether a listing of the
        // board happens to include it yet. See the module documentation.
        // An existing issue is never moved; a new one is created where the rule says — and
        // knowing where is what lets the board's fields and that repository's id be read
        // together, before anything below needs either.
        let destination = match target {
            Some(target) => {
                WriteDestination::Existing(Box::new(self.bound_item(target).await?.ok_or_else(
                    || SourceError::Refused {
                        message: format!("GitHub destination item {} was not found", target.0),
                    },
                )?))
            }
            None => {
                let target = self.creation_target(incoming).await?;
                self.creation_context(&target, incoming).await?;
                WriteDestination::New(target)
            }
        };
        let existing = destination.existing();
        let board = self
            .fields_for(
                existing,
                incoming.metadata.contains_key(ORIGIN_KEY),
                incoming.written.status().is_some(),
                incoming
                    .priority
                    .is_some_and(|priority| priority != Priority::None),
            )
            .await?;
        let status_target = incoming
            .written
            .work_status()
            .map(|(kind, status)| self.resolved_target(kind, status.category))
            .transpose()?;
        let column = match (incoming.written.work_status(), status_target.as_ref()) {
            (Some((kind, status)), Some(target)) => {
                self.column_for(&board.fields, kind, status.category, target)?
            }
            _ => None,
        };
        // Resolved before anything is created, for the reason the column above is: a
        // priority this board has no option for is refused while nothing has been written.
        let priority_write = match incoming.priority {
            Some(priority) => self.priority_write(&board.fields, existing, priority)?,
            None => None,
        };
        // And for the reason the priority is: a value no projected field can hold, or a board
        // that cannot hold the field, is refused while nothing has been written.
        let (projections, projected) = self.projection_writes(
            &board.fields,
            existing.map(|item| &item.projected),
            incoming.metadata,
            None,
        )?;
        let content_kind = existing.map_or(ContentKind::Issue, |item| item.content_kind);
        if content_kind == ContentKind::DraftIssue {
            if let (Some(StatusTarget::Terminal(_, _)), Some(status)) =
                (status_target.as_ref(), incoming.written.status())
            {
                return Err(self.closes_a_draft(status.category));
            }
            if incoming.parent.is_some() {
                return Err(SourceError::Refused {
                    message: "GitHub draft items cannot be a project's sub-issue".into(),
                });
            }
        }
        // The people it is to be assigned to are looked up — unless this process already has —
        // and a login GitHub does not know refused, while nothing has been written.
        if let Some(people) = incoming.assignees
            && existing.is_none_or(|item| {
                item.assignees
                    .as_deref()
                    .is_none_or(|held| !assignees::same_people(held, people))
            })
        {
            if content_kind == ContentKind::DraftIssue {
                return Err(self.assigns_a_draft());
            }
            self.resolve_people(people).await?;
            self.people_ids(
                people,
                &existing.map_or_else(
                    || format!("the new {}", incoming.title),
                    Resolved::written_name,
                ),
            )?;
        }
        match existing {
            Some(item) if content_kind == ContentKind::Issue => {
                if item.labels != incoming.labels {
                    return Err(SourceError::Refused {
                        message: "GitHub issue labels differ from the labels being written".into(),
                    });
                }
            }
            _ => {
                if !incoming.labels.is_empty() {
                    return Err(SourceError::Refused {
                        message: "GitHub items created by this destination carry no labels".into(),
                    });
                }
            }
        }

        // The repository the issue really lives in is what the slot below is written against,
        // so a single entry that is where the issue is created travels as no key at all, and
        // the read side derives it back from the issue.
        let own_repository = match &destination {
            WriteDestination::Existing(item) => item.own_repository.clone(),
            WriteDestination::New(target) => Some(
                Repository::try_from(target.origin())
                    .map_err(|message| SourceError::Config { message })?,
            ),
        };
        let (native, fallback) = self
            .partition_edges(
                incoming.written.kind(),
                content_kind,
                existing.and_then(|item| item.blocked_by.as_deref()),
                depends_on,
            )
            .await?;
        let mut slot = slot_metadata(incoming, own_repository.as_ref(), &fallback);
        let content = match incoming.assets {
            Some(assets) => {
                let uploads = self.upload_assets(own_repository.as_ref(), assets).await?;
                let rewritten = onetaskgraph_plugin_api::serve_asset_references(
                    incoming.content.unwrap_or_default(),
                    &mut slot,
                    &uploads,
                );
                incoming.content.map(|_| rewritten)
            }
            None => incoming.content.map(str::to_owned),
        };
        // A write over an existing item keeps its `## Links` section, re-rendered from the
        // links it holds.
        let content = content.map(|content| {
            onetaskgraph_plugin_api::relinked(
                &content,
                existing.and_then(|item| item.body.as_deref()),
                incoming.metadata,
            )
        });
        let marked = content.as_deref().map(|content| self.marked(content));
        let body = compose_body(marked.as_deref(), &slot)?;
        // Read before anything is created, for the reason the field below is: a value
        // this destination cannot store has to refuse, and refusing after `createIssue`
        // would leave an issue behind that nothing asked for. The engine writes a
        // qualified id here; a caller handing this key anything else is told so rather
        // than having it silently stored as no origin at all.
        // llmlint: ignore[boundary_inputs_validated, changed_behavior_has_e2e] The qualified id's syntax is the engine's and not this plugin's to police: `GlobalId` is deliberately absent from the contract crate because a plugin never sees a qualified id (AGENTS.md), no plugin crate may depend on the engine to parse one, and `docs/metadata.md` says the contents of this key are what no plugin constructs or interprets. What this boundary owns is whether the value is a string its text field can hold, and that is what it checks.
        let origin = match incoming.metadata.get(ORIGIN_KEY) {
            None => "",
            Some(Value::String(origin)) => origin.as_str(),
            Some(other) => {
                return Err(SourceError::Refused {
                    message: format!(
                        "{ORIGIN_KEY} holds a qualified id spelled as a string, and this item's \
                         is {other}"
                    ),
                });
            }
        };
        // Resolved before anything is created: a board that cannot carry the copy origin
        // has to refuse the write, and refusing it after `createIssue` would leave an
        // issue behind that nothing asked for.
        let origin_field = match Board::field(&board.fields, ORIGIN_FIELD)? {
            Some(field) => {
                if required_str(field, "__typename")? != "ProjectV2Field" {
                    return Err(SourceError::Refused {
                        message: format!(
                            "GitHub board source-owned {ORIGIN_FIELD} field is not a text field"
                        ),
                    });
                }
                Some(required_str(field, "id")?.to_owned())
            }
            None if incoming.metadata.contains_key(ORIGIN_KEY) => {
                return Err(SourceError::Refused {
                    message: format!(
                        "GitHub board has no source-owned {ORIGIN_FIELD} text field, and the \
                         item carries {ORIGIN_KEY}; add a text field named {ORIGIN_FIELD} to \
                         the board"
                    ),
                });
            }
            None => None,
        };

        Ok(Prepared {
            origin: origin.to_owned(),
            destination,
            board,
            status_target,
            column,
            priority_write,
            projections,
            projected,
            content_kind,
            own_repository,
            native,
            slot,
            content,
            body,
            origin_field,
        })
    }

    /// Everything one write does once the item it writes exists: its board fields, its
    /// relationships and — for an item that was already there — its content last, then what this
    /// process remembers of it.
    ///
    /// While this source is holding writes back — see [`TaskSource::begin_deferred_writes`] — an
    /// item this command created has its fields, its closing and its edges held for
    /// [`TaskSource::finish_writes`] rather than sent here.
    async fn complete_write(
        &self,
        incoming: &Incoming<'_>,
        prepared: Prepared,
        landed: Landed,
    ) -> Result<onetaskgraph_plugin_api::AssetsWritten, SourceError> {
        let Prepared {
            destination,
            board,
            status_target,
            column,
            priority_write,
            projections,
            projected,
            content_kind,
            own_repository,
            native,
            slot,
            content,
            body,
            origin,
            origin_field,
        } = prepared;
        let existing = destination.existing();
        let origin = origin.as_str();
        let Landed {
            content_id,
            item_id,
            url,
            number,
        } = landed;

        // Only what differs from the item as this write read it is sent: a priority the item
        // already holds is not, so an unchanged re-copy writes no field and a person's edit of
        // it since that read is left alone.
        let priority_write = priority_write.filter(|_| {
            !existing.is_some_and(|item| {
                incoming.priority.map(HeldPriority::Read) == Some(item.priority.clone())
            })
        });
        let written_option = column.as_ref().map(|(_, _, name)| name.clone());
        let column = column
            .filter(|(_, _, name)| existing.is_none_or(|item| item.option.as_ref() != Some(name)))
            .map(|(field, option, _)| (field, option));
        // Creating an item here is several calls — `createIssue`, which files it on the
        // board, then its board fields, the parent and the dependencies — and GitHub can fail
        // at any of them. Everything this source can refuse *before* the first of those is
        // already checked above, so what is left is GitHub itself failing part way. When it
        // does over an item this call created, the issue is taken back: a write that
        // refused must not leave an item behind that nobody asked for, and one that does
        // makes the retry create a second.
        // Whether the board-field write carrying a moved origin was answered as landing whole.
        // When it was refused, GitHub does not say which of its fields ran before the one that
        // failed, so the origin may or may not have moved.
        let mut origin_landed = false;
        // An item this command created while writes are held back has every part that can land
        // later held for `finish_writes` — its own create filed it on the board and under its
        // project already — and so does a later write of it in the same command.
        let holding = (existing.is_none() && self.deferring()?) || self.holds_back(&content_id)?;
        let landed = if holding {
            self.hold_back(
                &board.id,
                &content_id,
                existing,
                origin_field.as_deref(),
                origin,
                column,
                status_target.as_ref(),
                priority_write.as_ref(),
                &projections,
                &native,
            )
        } else if let Some(item_id) = item_id.as_deref() {
            self.finish_write(
                board.id.as_str(),
                incoming,
                &content_id,
                item_id,
                content_kind,
                existing,
                origin_field.as_deref(),
                origin,
                column,
                status_target.as_ref(),
                priority_write.as_ref(),
                &projections,
                &native,
                &mut origin_landed,
            )
            .await
        } else {
            // An issue a create filed on the board itself has no board item this process
            // knows until `finish_writes` reads it, so nothing but a held-back write may land
            // on it before then.
            Err(SourceError::Malformed {
                message: format!(
                    "issue {} was written before the board item GitHub filed it as was read",
                    content_id.0
                ),
            })
        };
        // An existing item's title, body and state go last, in one `updateIssue`, once its board
        // fields and its relationships have landed: a refusal of any of those then leaves its
        // body — and the metadata slot inside it — exactly as it stood.
        let landed = match (landed, existing) {
            (Ok(()), Some(item)) => {
                self.update_existing(item, incoming, &body, status_target.as_ref())
                    .await
            }
            (landed, _) => landed,
        };
        if let Err(error) = landed {
            match existing {
                // Best effort, and the write's own failure is what the caller is told: a
                // refusal naming the tidy-up would hide why the write failed at all.
                None => {
                    let _ = self.delete_issue(&content_id).await;
                }
                // The origin field is the one piece of an existing item's metadata written
                // before its body, so a write refused after it puts it back as it was. When
                // that is refused too, the write's own failure is still what the caller is
                // told — with what it left behind added, because the item's metadata is then
                // not as it stood and a caller retrying has to know which key moved.
                Some(item) => {
                    let before = item.origin.as_deref().unwrap_or("");
                    if let Some(field) = origin_field.as_deref()
                        && before != origin
                        && let Err(restore) = self
                            .set_item_field(
                                board.id.as_str(),
                                &item.item_id,
                                field,
                                json!({"text": before}),
                            )
                            .await
                    {
                        let left = if origin_landed {
                            format!(
                                "its {ORIGIN_KEY} was moved to {origin:?} before that and could \
                                 not be put back to {before:?} ({restore}), so item {} still \
                                 holds {origin:?} there",
                                item.id.0
                            )
                        } else {
                            format!(
                                "the refused write carried its {ORIGIN_KEY} from {before:?} to \
                                 {origin:?}, GitHub does not say whether that part of it ran, \
                                 and putting it back to {before:?} was refused ({restore}), so \
                                 item {} holds {origin:?} or {before:?} there",
                                item.id.0
                            )
                        };
                        return Err(noting(
                            error,
                            &format!(
                                "; {left}; next: set {ORIGIN_KEY} on it back to {before:?}, or \
                                 run the write again"
                            ),
                        ));
                    }
                }
            }
            return Err(error);
        }

        let written_status = match (incoming.written.work_status(), status_target.as_ref()) {
            (Some((kind, _)), Some(StatusTarget::Terminal(_, reason))) => {
                self.statuses
                    .status(kind, written_option.as_deref(), true, Some(reason.reason()))
            }
            (Some((kind, _)), Some(StatusTarget::Column(_))) => {
                self.statuses
                    .status(kind, written_option.as_deref(), false, None)
            }
            (Some((_, status)), _) => status.clone(),
            (None, _) => Status {
                category: StatusCategory::Unknown,
                name: "Open".to_owned(),
            },
        };

        // So the rest of this command reads what it just did rather than what the board
        // said before it. See `remember_written` for which half takes it.
        let remembered = Resolved {
            // Empty for an issue whose board item is not read yet: `finish_writes` reads it
            // and `learn_item` puts it here.
            // llmlint: ignore[invalid_states_unrepresentable] `Resolved` is serialised into the cache's stored write context, so its field type is a stored shape this change may not move; the empty id lives only until `finish_writes` reads the board item in this same command, and nothing addresses it before then — `complete_write` refuses any write naming no board item.
            item_id: item_id.unwrap_or_default(),
            id: content_id.clone(),
            content_kind,
            kind: incoming.written.kind(),
            title: incoming.title.to_owned(),
            // The visible half of the body this write composed, split back off it the
            // way a read splits it — so what this record reports is what a read of the
            // same issue reports, rather than the person's text with the metadata slot
            // still on the end of it.
            body: self.visible(body.clone())?.0,
            raw_body: body.clone(),
            // A document has no status of its own; what it reads back as is whatever
            // the issue's own state says, which is what a re-read reports.
            status: written_status,
            option: written_option.or_else(|| existing.and_then(|item| item.option.clone())),
            priority: match incoming.priority {
                Some(priority) => HeldPriority::Read(priority),
                None => existing.map_or(HeldPriority::Read(Priority::None), |item| {
                    item.priority.clone()
                }),
            },
            // What `state_input` asked for: closed for a terminal target, open for any other
            // status, and the issue's own state left as it was by a document write.
            closed: content_kind == ContentKind::Issue
                && match status_target.as_ref() {
                    Some(StatusTarget::Terminal(_, _)) => true,
                    Some(_) => false,
                    None => existing.is_some_and(|item| item.closed),
                },
            reason: match status_target.as_ref() {
                Some(StatusTarget::Terminal(_, reason)) if content_kind == ContentKind::Issue => {
                    Some(reason.reason().to_owned())
                }
                Some(_) => None,
                None => existing.and_then(|item| item.reason.clone()),
            },
            delivers: incoming.delivers.to_vec(),
            delivered_by: incoming.delivered_by.to_vec(),
            labels: incoming.labels.to_vec(),
            parent: incoming.parent.cloned(),
            issue_parent: None,
            origin: (!origin.is_empty()).then(|| origin.to_owned()),
            projected,
            number,
            // In the update path this is the item's own url, read off `existing` where the
            // record above was bound, so one expression serves both halves.
            url,
            created_at: existing.and_then(|item| item.created_at),
            updated_at: existing.and_then(|item| item.updated_at),
            // What this write moved is GitHub's to say, so a record of it gives no token until
            // it is read again.
            item_updated_at: None,
            comment_count: None,
            relationship_events: None,
            served_at: None,
            facets: None,
            own_repository,
            repositories: incoming.repositories.to_vec(),
            classification: incoming.classification,
            slot,
            board_id: Some(board.id.as_str().to_owned()),
            fields: board
                .fields
                .get("nodes")
                .and_then(Value::as_array)
                .cloned()
                .unwrap_or_default(),
            board_fields: Some(board.fields.clone()),
            // What this write left the relationship holding is known by id alone, and a
            // later read of its edges needs each far end's kind, so it reads them again.
            blocked_by: None,
            // The people it named, which its own answer was held to; none named keeps the
            // item's, and creates a task of an opted-in source unassigned.
            assignees: match incoming.assignees {
                Some(people) => Some(people.to_vec()),
                None => match existing {
                    Some(item) => item.assignees.clone(),
                    None => (self.assignees
                        && matches!(incoming.written, Written::Work(ItemKind::Task, _)))
                    .then(Vec::new),
                },
            },
        };
        self.remember_written(remembered, existing.is_none())?;
        Ok(onetaskgraph_plugin_api::AssetsWritten {
            id: content_id,
            content,
        })
    }

    /// Everything a write does after the item exists: its board fields, its parent, and
    /// its dependencies.
    ///
    /// Split out of `write_item` so there is one place a failure past the point of no
    /// return is caught, rather than a tidy-up repeated at each `?` above.
    // llmlint: ignore[suppressions_justified] This is the tail of `write_item` lifted out
    // so there is one place a failure past the point of no return is caught, and its
    // arguments are exactly the values that tail already had in scope. Bundling them into a
    // struct would describe no concept — it would be "the arguments of this function" — and
    // would put the whole of `write_item`'s locals behind one more indirection.
    #[allow(clippy::too_many_arguments)]
    async fn finish_write(
        &self,
        board_id: &str,
        incoming: &Incoming<'_>,
        content_id: &NativeId,
        item_id: &str,
        content_kind: ContentKind,
        existing: Option<&Resolved>,
        origin_field: Option<&str>,
        origin: &str,
        column: Option<(String, String)>,
        status_target: Option<&StatusTarget>,
        priority: Option<&PriorityWrite>,
        projections: &[ProjectionWrite],
        native: &[String],
        origin_landed: &mut bool,
    ) -> Result<(), SourceError> {
        let (fields, clears) = field_writes(
            existing,
            origin_field,
            origin,
            column,
            priority,
            projections,
        );
        self.set_item_fields(board_id, item_id, &fields, &clears)
            .await?;
        *origin_landed = true;

        // An existing issue closes in the `updateIssue` its write ends with; one created just
        // now closes here, once its option is selected.
        if existing.is_none()
            && content_kind == ContentKind::Issue
            && matches!(status_target, Some(StatusTarget::Terminal(_, _)))
        {
            self.update_content(
                ContentKind::Issue,
                content_id,
                json!({"stateInput":state_input(status_target)}),
            )
            .await?;
        }

        if content_kind == ContentKind::Issue {
            self.reparent(
                existing.and_then(|item| item.parent.clone()),
                content_id,
                incoming.parent,
            )
            .await?;
            // A document takes part in no dependency graph, so writing one neither reads
            // nor changes the issue's own `blockedBy` relationships. Reconciling them
            // against the empty list a document write carries would *delete* whatever
            // relationships a person had made on that issue, which is a write nobody
            // asked for.
            if incoming.written.kind() != BoardKind::Document {
                let issue = match existing {
                    Some(item) => Issue::Existing(item.blocked_by.as_deref()),
                    None => Issue::Created,
                };
                self.reconcile_blocked_by(content_id, native, issue).await?;
            }
        }
        Ok(())
    }

    /// Whether this source is holding writes back — see [`TaskSource::begin_deferred_writes`].
    fn deferring(&self) -> Result<bool, SourceError> {
        Ok(self.deferred()?.is_some())
    }

    /// Whether `content` is an item this command created whose parts are held back.
    fn holds_back(&self, content: &NativeId) -> Result<bool, SourceError> {
        Ok(self
            .deferred()?
            .as_ref()
            .is_some_and(|deferred| deferred.holds(content)))
    }

    fn deferred(&self) -> Result<std::sync::MutexGuard<'_, Option<Deferred>>, SourceError> {
        self.deferred.lock().map_err(|_| SourceError::Unavailable {
            message: "this source's record of the writes it is holding back was left \
                      inconsistent by an earlier failure; next: run the command again"
                .into(),
        })
    }

    /// Creates one issue filed on the board and under its project by the `createIssue` itself —
    /// `projectV2Ids` naming the board and `parentIssueId` the project — which is what a copy
    /// creates with: one request where adding it to the board and filing it under its project
    /// were two more. GitHub adds the board item after answering, so its id is not known here;
    /// [`TaskSource::finish_writes`] reads it.
    async fn create_filed_issue(
        &self,
        board_id: &str,
        repository: &RepositoryTarget,
        incoming: &Incoming<'_>,
        body: &Option<String>,
    ) -> Result<Landed, SourceError> {
        let repository_id = self.repository_id(repository, incoming).await?;
        let mut input = filed_input(&repository_id, board_id, incoming, body);
        self.assigning(&mut input, incoming)?;
        let data = match self
            .graphql(graphql::CREATE_ISSUE, json!({"input": &input}))
            .await
        {
            Ok(data) => data,
            Err(error) => return Err(self.refusing_people(&input, error)),
        };
        let created = data
            .pointer("/createIssue/issue")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub issue creation returned no issue".into(),
            })?;
        let landed = self.created_landed(created).await?;
        // An issue GitHub created assigned otherwise than asked is refused and taken back; a
        // delete GitHub refuses too is answered beside the refusal, naming the issue left.
        if let Err(error) = self.assigned_as_asked(&input, Some(created)) {
            return Err(self.taken_back(&landed.content_id, created, error).await);
        }
        Ok(landed)
    }

    /// Keep the address `createIssue` answered for the issue `content`, for
    /// [`TaskSource::written_location`].
    fn note_written(&self, content: &NativeId, url: Option<&str>) {
        if let Some(url) = url {
            self.written
                .lock()
                .unwrap_or_else(std::sync::PoisonError::into_inner)
                .insert(content.clone(), url.to_owned());
        }
    }

    /// What one `createIssue` answered, as the item it created — or, when its number cannot be
    /// read, the refusal, after taking the issue back.
    async fn created_landed(&self, created: &Value) -> Result<Landed, SourceError> {
        let content_id = NativeId(required_nonblank_str(created, "id")?.to_owned());
        let url = optional_str(created, "url")?.map(str::to_owned);
        let number = match created_issue_number(created) {
            Ok(number) => number,
            Err(error) => {
                let _ = self.delete_issue(&content_id).await;
                return Err(error);
            }
        };
        self.note_written(&content_id, url.as_deref());
        Ok(Landed {
            content_id,
            item_id: None,
            url,
            number,
        })
    }

    /// Hold back everything a write of the created item `content` would send after creating it,
    /// for [`TaskSource::finish_writes`]: the same field values, closing and edges
    /// `finish_write` sends, the parent excepted, which its create already named. A later write
    /// of the same item replaces what it changes.
    // llmlint: ignore[suppressions_justified] The arguments are `finish_write`'s, which this holds back instead of sending, and they are exactly the values `complete_write` has in scope; bundling them would describe no concept.
    #[allow(clippy::too_many_arguments)]
    fn hold_back(
        &self,
        board_id: &BoardId,
        content: &NativeId,
        existing: Option<&Resolved>,
        origin_field: Option<&str>,
        origin: &str,
        column: Option<(String, String)>,
        status_target: Option<&StatusTarget>,
        priority: Option<&PriorityWrite>,
        projections: &[ProjectionWrite],
        native: &[String],
    ) -> Result<(), SourceError> {
        let status = column.map(|(field, option)| {
            let name = self
                .fields_cache()
                .ok()
                .and_then(|fields| {
                    fields.as_ref().and_then(|board| {
                        option_name(&board.fields, &field, &option).map(str::to_owned)
                    })
                })
                .unwrap_or_default();
            (field, option, name)
        });
        let (fields, clears) =
            field_writes(existing, origin_field, origin, None, priority, projections);
        let close = match status_target {
            Some(StatusTarget::Terminal(_, reason)) => Some(*reason),
            _ => None,
        };
        let mut deferred = self.deferred()?;
        let deferred = deferred.get_or_insert_with(Deferred::default);
        if let Deferred::Nothing = deferred {
            *deferred = Deferred::Held {
                board: board_id.clone(),
                created: Vec::new(),
            };
        }
        let Deferred::Held { created, .. } = deferred else {
            unreachable!("held back on the line above");
        };
        let held = match created.iter().position(|held| held.content == *content) {
            Some(at) => &mut created[at],
            None => {
                created.push(HeldBack {
                    content: content.clone(),
                    fields: Vec::new(),
                    status: None,
                    clears: Vec::new(),
                    close: None,
                    edges: Vec::new(),
                });
                created.last_mut().expect("just held")
            }
        };
        for (field, value) in fields {
            held.fields.retain(|(held, _)| *held != field);
            held.clears.retain(|held| *held != field);
            held.fields.push((field, value));
        }
        for field in clears {
            held.fields.retain(|(held, _)| held != field);
            if !held.clears.iter().any(|held| held == field) {
                held.clears.push(field.to_owned());
            }
        }
        if status.is_some() {
            held.status = status;
        }
        if close.is_some() {
            held.close = close;
        }
        // Nothing of an item created in this command has been sent but its create, so the edges
        // it is to have are exactly the ones its latest write names.
        held.edges = native.iter().map(|far| NativeId(far.clone())).collect();
        Ok(())
    }

    /// Create every task of `writes`, together, filed on the board and under each one's
    /// project: [`graphql::CREATE_ISSUES`], [`MUTATION_BATCH`] to a request.
    ///
    /// Each is prepared as [`Self::write_item`] prepares one — every refusal a write owes made
    /// before the first request — then created, then completed with its parts held back. A
    /// request refused part way leaves the issues it did create, which are taken back, with the
    /// rest of this call's, before the refusal is answered.
    async fn create_together(
        &self,
        writes: &[ItemWrite<Task>],
    ) -> Result<Vec<NativeId>, SourceError> {
        let incomings: Vec<Incoming<'_>> = writes
            .iter()
            .map(|write| self.task_incoming(write, None))
            .collect::<Result<_, _>>()?;
        // Every person these creates assign, looked up together rather than one write at a time.
        let people: Vec<String> = writes
            .iter()
            .filter_map(|write| write.item.assignees.clone())
            .flatten()
            .collect();
        self.resolve_people(&people).await?;
        let mut prepared = Vec::with_capacity(writes.len());
        for (incoming, write) in incomings.iter().zip(writes) {
            if write.target.is_some() {
                return Err(SourceError::Malformed {
                    message: "a write naming its target was handed to a batch of creates".into(),
                });
            }
            prepared.push(
                self.prepare_write(incoming, None, &write.depends_on)
                    .await?,
            );
        }
        let mut inputs = Vec::with_capacity(prepared.len());
        for (incoming, prepared) in incomings.iter().zip(&prepared) {
            let WriteDestination::New(target) = &prepared.destination else {
                return Err(SourceError::Malformed {
                    message: "an update was decided for a write handed to a batch of creates"
                        .into(),
                });
            };
            let repository_id = self.repository_id(target, incoming).await?;
            let mut input = filed_input(
                &repository_id,
                prepared.board.id.as_str(),
                incoming,
                &prepared.body,
            );
            self.assigning(&mut input, incoming)?;
            inputs.push(input);
        }
        let mut landed: Vec<Landed> = Vec::with_capacity(inputs.len());
        let refused = 'create: {
            for chunk in inputs.chunks(MUTATION_BATCH) {
                let mut variables = serde_json::Map::new();
                for slot in 0..MUTATION_BATCH {
                    variables.insert(
                        format!("create{slot}"),
                        chunk.get(slot).cloned().unwrap_or_else(|| chunk[0].clone()),
                    );
                    variables.insert(format!("send{slot}"), json!(slot < chunk.len()));
                }
                let data = match self
                    .graphql(graphql::CREATE_ISSUES, Value::Object(variables))
                    .await
                {
                    Ok(data) => data,
                    Err(error) => break 'create Some(error),
                };
                let mut refusal = data.get(ANSWER_ERRORS).map(|errors| {
                    let said: Vec<&str> = errors
                        .as_array()
                        .into_iter()
                        .flatten()
                        .filter_map(|error| error.get("message").and_then(Value::as_str))
                        .collect();
                    SourceError::Refused {
                        message: if said.is_empty() {
                            "GitHub returned GraphQL errors".to_owned()
                        } else {
                            said.join("; ")
                        },
                    }
                });
                for (slot, input) in chunk.iter().enumerate() {
                    match data
                        .pointer(&format!("/create{slot}/issue"))
                        .filter(|value| !value.is_null())
                    {
                        Some(created) => match self.created_landed(created).await {
                            Ok(landed_one) => {
                                landed.push(landed_one);
                                // Taken back with the rest, as any refusal here is.
                                if let Err(error) = self.assigned_as_asked(input, Some(created)) {
                                    refusal.get_or_insert(error);
                                }
                            }
                            Err(error) => {
                                refusal.get_or_insert(error);
                            }
                        },
                        // An alias GitHub refused answers nothing, and the refusal says why.
                        None => {
                            refusal.get_or_insert(SourceError::Malformed {
                                message: "GitHub issue creation returned no issue".into(),
                            });
                        }
                    }
                }
                if let Some(error) = refusal {
                    break 'create Some(error);
                }
            }
            None
        };
        if let Some(error) = refused {
            // None of these ids reaches the caller, so an issue GitHub will not take back is
            // named here or by nobody.
            let ids: Vec<&NativeId> = landed.iter().map(|created| &created.content_id).collect();
            let left = self.delete_issues(&ids).await;
            let left: Vec<&str> = left.iter().map(|id| id.0.as_str()).collect();
            return Err(left_behind(error, &left));
        }
        let created: Vec<NativeId> = landed
            .iter()
            .map(|landed| landed.content_id.clone())
            .collect();
        let mut ids = Vec::with_capacity(landed.len());
        for ((incoming, prepared), landed) in incomings.iter().zip(prepared).zip(landed) {
            match self.complete_write(incoming, prepared, landed).await {
                Ok(written) => ids.push(written.id),
                // llmlint: ignore[changed_behavior_has_e2e] While writes are held back, which is the only time a batch is created, completing a created item only holds its parts back, and that fails on nothing but a poisoned lock, which no journey can cause; the issues are taken back regardless, because the caller is never handed their ids.
                Err(error) => {
                    let ids: Vec<&NativeId> = created.iter().collect();
                    let left = self.delete_issues(&ids).await;
                    let left: Vec<&str> = left.iter().map(|id| id.0.as_str()).collect();
                    return Err(left_behind(error, &left));
                }
            }
        }
        Ok(ids)
    }

    /// Hold back the update of the existing issue `issue` — `members`, what its own
    /// [`graphql::UPDATE_ISSUE`] would send beside its id — while writes are being held back — so the updates a copy makes one after
    /// another go to GitHub as one request. Whatever this source held of the issue is dropped
    /// as the mutation itself would drop it. Answers whether it was held: never while a write
    /// the store's cache expects runs, which records each mutation's own answer, and never for
    /// an issue an update is already held for, which that one is sent ahead of.
    fn hold_update(&self, issue: &NativeId, members: &Value) -> Result<bool, SourceError> {
        if self.expected.journaling() || self.deferred()?.is_none() {
            return Ok(false);
        }
        let mut held = self
            .held_updates
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "issue updates held back were left inconsistent; run the command again"
                    .into(),
            })?;
        if held.iter().any(|earlier| earlier.issue == *issue) {
            return Ok(false);
        }
        let named = |value: &str| issue.0 == value;
        self.resolved_cache()?
            .retain(|id, item| !named(&id.0) && !named(&item.item_id));
        self.read_ahead()?.retain(|id, item| {
            !named(&id.0) && !item.as_ref().is_some_and(|item| named(&item.item_id))
        });
        self.carried()?
            .retain(|id, item| !named(&id.0) && !named(&item.item_id));
        held.push(HeldUpdate::new(issue, members));
        Ok(true)
    }

    /// Send every update [`Self::hold_update`] held back: one alone as [`graphql::UPDATE_ISSUE`]
    /// sends it, several as aliased `updateIssue`s of [`graphql::UPDATE_ISSUES`],
    /// [`MUTATION_BATCH`] to a request. Each counts one write, as it would alone.
    async fn send_held_updates(&self) -> Result<(), SourceError> {
        let held = std::mem::take(&mut *self.held_updates.lock().map_err(|_| {
            SourceError::Unavailable {
                message: "issue updates held back were left inconsistent; run the command again"
                    .into(),
            }
        })?);
        if let [one] = held.as_slice() {
            return self
                .send_update(
                    graphql::UPDATE_ISSUE,
                    "/updateIssue/issue",
                    &one.issue,
                    one.input.clone(),
                )
                .await;
        }
        for chunk in held.chunks(MUTATION_BATCH) {
            let mut variables = serde_json::Map::new();
            for slot in 0..MUTATION_BATCH {
                variables.insert(
                    format!("update{slot}"),
                    chunk.get(slot).unwrap_or(&chunk[0]).input.clone(),
                );
                variables.insert(format!("send{slot}"), json!(slot < chunk.len()));
            }
            let data = match self
                .graphql(graphql::UPDATE_ISSUES, Value::Object(variables))
                .await
            {
                Ok(data) => data,
                Err(error) => {
                    for held in chunk {
                        self.forget_people(&held.input);
                    }
                    return Err(error);
                }
            };
            let inputs: Vec<&Value> = chunk.iter().map(|held| &held.input).collect();
            check_aliased_answers(
                graphql::UPDATE_ISSUES,
                &data,
                "update",
                &inputs,
                &[("issue", "id")],
            )?;
            for (slot, held) in chunk.iter().enumerate() {
                self.assigned_as_asked(&held.input, data.pointer(&format!("/update{slot}/issue")))?;
            }
        }
        Ok(())
    }

    /// Send everything held back since [`TaskSource::begin_deferred_writes`]: one read of the
    /// created items' board items, then their field values and edges together in requests of
    /// [`WRITE_BATCH`], then their closes in requests of [`MUTATION_BATCH`]. Anything refused
    /// takes back every issue held, [`DELETE_BATCH`] to a request, and answers the refusal.
    async fn send_held_back(&self) -> Result<(), SourceError> {
        let Some(Deferred::Held { board, created }) = self.deferred()?.take() else {
            return Ok(());
        };
        if created.is_empty() {
            return Ok(());
        }
        let sent = self.send_deferred(board.as_str(), &created).await;
        if let Err(error) = sent {
            // Best effort, and the failure is what the caller is told, as a write of one item
            // that fails part way takes back what it created: the copy that held these back
            // journalled every one, so one this cannot take back its undo tries again, and says
            // so when it cannot either.
            let ids: Vec<&NativeId> = created.iter().map(|held| &held.content).collect();
            let _ = self.delete_issues(&ids).await;
            return Err(error);
        }
        Ok(())
    }

    /// The requests [`Self::send_held_back`] makes.
    async fn send_deferred(&self, board: &str, created: &[HeldBack]) -> Result<(), SourceError> {
        let items = self.created_board_items(board, created).await?;
        let mut fields: Vec<Value> = Vec::new();
        let mut closes: Vec<Value> = Vec::new();
        let mut edges: Vec<Value> = Vec::new();
        for held in created {
            let (item, status) = &items[&held.content];
            let mut values = held.fields.clone();
            // A status the board already shows — set by a workflow a person turned on to file
            // new items at it — is not written again.
            if let Some((field, option, name)) = &held.status
                && status.as_deref() != Some(name.as_str())
            {
                values.insert(0, (field.clone(), json!({"singleSelectOptionId": option})));
            }
            for (field, value) in values {
                fields.push(json!({"projectId":board,"itemId":item,"fieldId":field,"value":value}));
            }
            if let Some(reason) = held.close {
                closes.push(json!({"id": held.content.0, "stateInput": closed_input(reason)}));
            }
            for far in &held.edges {
                edges.push(json!({"issueId": held.content.0, "blockingIssueId": far.0}));
            }
            if !held.clears.is_empty() {
                let clears: Vec<&str> = held.clears.iter().map(String::as_str).collect();
                self.set_item_fields(board, item, &[], &clears).await?;
            }
        }
        // Each alias answers with what it wrote, which is what its input named: the board item a
        // field value is set on, the issue closed, the issue blocked and the one blocking it.
        let field_answer: &[(&str, &str)] = &[("projectV2Item", "itemId")];
        let close_answer: &[(&str, &str)] = &[("issue", "id")];
        let block_answer: &[(&str, &str)] =
            &[("issue", "issueId"), ("blockingIssue", "blockingIssueId")];
        // Every field value, then every edge, as one ordered list sent `WRITE_BATCH` aliases to a
        // request: a request holding both kinds is the one document carrying both, so a plan's
        // values and edges cost one request where they fit in one.
        let writes: Vec<HeldWrite<'_>> = fields
            .iter()
            .map(HeldWrite::Field)
            .chain(edges.iter().map(HeldWrite::Edge))
            .collect();
        for chunk in writes.chunks(WRITE_BATCH) {
            let values: Vec<&Value> = chunk
                .iter()
                .filter_map(|write| match write {
                    HeldWrite::Field(input) => Some(*input),
                    HeldWrite::Edge(_) => None,
                })
                .collect();
            let blocks: Vec<&Value> = chunk
                .iter()
                .filter_map(|write| match write {
                    HeldWrite::Edge(input) => Some(*input),
                    HeldWrite::Field(_) => None,
                })
                .collect();
            let (document, sends) = match (values.is_empty(), blocks.is_empty()) {
                (false, false) => (graphql::WRITE_FIELDS_AND_EDGES, ["sendField", "sendBlock"]),
                (false, true) => (graphql::WRITE_FIELD_VALUES, ["send", "send"]),
                _ => (graphql::ADD_BLOCKED_BYS, ["send", "send"]),
            };
            let mut variables = serde_json::Map::new();
            let kinds = [
                ("field", &values, sends[0], field_answer),
                ("block", &blocks, sends[1], block_answer),
            ];
            for (prefix, inputs, send, _) in &kinds {
                if inputs.is_empty() {
                    continue;
                }
                for slot in 0..WRITE_BATCH {
                    variables.insert(
                        format!("{prefix}{slot}"),
                        (*inputs.get(slot).unwrap_or(&inputs[0])).clone(),
                    );
                    variables.insert(format!("{send}{slot}"), json!(slot < inputs.len()));
                }
            }
            let data = self.graphql(document, Value::Object(variables)).await?;
            for (prefix, inputs, _, answers) in kinds {
                check_aliased_answers(document, &data, prefix, inputs, answers)?;
            }
        }
        for chunk in closes.chunks(MUTATION_BATCH) {
            let mut variables = serde_json::Map::new();
            for slot in 0..MUTATION_BATCH {
                variables.insert(
                    format!("close{slot}"),
                    chunk.get(slot).cloned().unwrap_or_else(|| chunk[0].clone()),
                );
                variables.insert(format!("send{slot}"), json!(slot < chunk.len()));
            }
            let data = self
                .graphql(graphql::CLOSE_ISSUES, Value::Object(variables))
                .await?;
            let inputs: Vec<&Value> = chunk.iter().collect();
            check_aliased_answers(graphql::CLOSE_ISSUES, &data, "close", &inputs, close_answer)?;
        }
        for held in created {
            self.learn_item(&held.content, &items[&held.content].0)?;
        }
        Ok(())
    }

    /// The board item each of `created` is on this board, with its `Status` as the board set it:
    /// [`graphql::CREATED_BOARD_ITEMS`], read again — after a wait of a second, then two, four,
    /// eight and sixteen — for only those GitHub has not filed yet.
    ///
    /// **Membership is never assumed.** `createIssue`'s `projectV2Ids` files an issue some time
    /// after GitHub answered the create — the feasibility spike saw 0.45 to 1.84 seconds, and no
    /// create answer carrying the item — so only a read says it is there. One still absent once
    /// those reads are spent is filed with [`graphql::ADD_TO_BOARD`], one write each; GitHub
    /// filing it meanwhile refuses that "Content already exists in this project", which says it
    /// is placed, and one more read finds the item rather than a second one being made.
    async fn created_board_items(
        &self,
        board: &str,
        created: &[HeldBack],
    ) -> Result<BTreeMap<NativeId, (String, Option<String>)>, SourceError> {
        let mut found: BTreeMap<NativeId, (String, Option<String>)> = BTreeMap::new();
        let unfiled = |found: &BTreeMap<NativeId, (String, Option<String>)>| -> Vec<NativeId> {
            created
                .iter()
                .map(|held| held.content.clone())
                .filter(|id| !found.contains_key(id))
                .collect()
        };
        let mut wait = Duration::from_secs(1);
        for attempt in 0..6 {
            let absent = unfiled(&found);
            if absent.is_empty() {
                return Ok(found);
            }
            if attempt > 0 {
                self.clock
                    .sleep_because(wait, pacing::cause::BOARD_FILING)
                    .await;
                wait = wait.saturating_mul(2);
            }
            self.read_created_items(board, &absent, &mut found).await?;
        }
        let mut placed_meanwhile = Vec::new();
        for id in unfiled(&found) {
            match self
                .graphql(
                    graphql::ADD_TO_BOARD,
                    json!({"input":{"projectId":board,"contentId":id.0}}),
                )
                .await
            {
                Ok(added) => {
                    let item = added
                        .pointer("/addProjectV2ItemById/item")
                        .filter(|value| !value.is_null())
                        .ok_or_else(|| SourceError::Malformed {
                            message: "GitHub board addition returned no project item".into(),
                        })?;
                    found.insert(id, (required_nonblank_str(item, "id")?.to_owned(), None));
                }
                Err(error) if already_on_the_board(&error) => placed_meanwhile.push(id),
                Err(error) => return Err(error),
            }
        }
        if !placed_meanwhile.is_empty() {
            self.read_created_items(board, &placed_meanwhile, &mut found)
                .await?;
        }
        let absent = unfiled(&found);
        if absent.is_empty() {
            return Ok(found);
        }
        let absent: Vec<&str> = absent.iter().map(|id| id.0.as_str()).collect();
        Err(SourceError::Unavailable {
            message: format!(
                "GitHub created {} filed on board {board} and, after 31 seconds and adding {} \
                 to it, still answers {} on no item of it: {}; next: run the copy again once \
                 GitHub has caught up",
                if absent.len() == 1 {
                    "an issue"
                } else {
                    "issues"
                },
                if absent.len() == 1 { "it" } else { "them" },
                if absent.len() == 1 { "it" } else { "them" },
                absent.join(", ")
            ),
        })
    }

    /// One [`graphql::CREATED_BOARD_ITEMS`] read of each of `ids`, [`COPY_DETAIL_BATCH`] to a
    /// request, adding to `found` the board item of every one GitHub has filed on `board`.
    async fn read_created_items(
        &self,
        board: &str,
        ids: &[NativeId],
        found: &mut BTreeMap<NativeId, (String, Option<String>)>,
    ) -> Result<(), SourceError> {
        for chunk in ids.chunks(COPY_DETAIL_BATCH) {
            let mut variables = serde_json::Map::new();
            for slot in 0..COPY_DETAIL_BATCH {
                variables.insert(
                    format!("id{slot}"),
                    json!(chunk.get(slot).unwrap_or(&chunk[0]).0),
                );
                variables.insert(format!("read{slot}"), json!(slot < chunk.len()));
            }
            variables.insert("boardItems".into(), json!(BOARD_ITEMS_PAGE_SIZE));
            variables.insert("statusName".into(), json!(STATUS_FIELD));
            let data = self
                .graphql(graphql::CREATED_BOARD_ITEMS, Value::Object(variables))
                .await?;
            for (slot, id) in chunk.iter().enumerate() {
                // An answer for another issue than the one asked about says nothing of
                // where that one is filed, and its board item is not this one's.
                let node = data.get(format!("i{slot}")).filter(|node| !node.is_null());
                let named = node.map(|node| node.get("id").and_then(Value::as_str));
                if let Some(named) = named
                    && named != Some(id.0.as_str())
                {
                    return Err(SourceError::Malformed {
                        message: format!(
                            "GitHub answered the board items of {} when asked for those of {}",
                            named.unwrap_or("an issue it did not name"),
                            id.0
                        ),
                    });
                }
                // An issue GitHub does not answer yet is not filed yet; one it answers is
                // read whole, its connection, each board item's board and its status, and
                // any of them unreadable is refused rather than read as not filed.
                let Some(node) = node else { continue };
                let malformed = |what: &str| SourceError::Malformed {
                    message: format!("GitHub answered the board items of {} with {what}", id.0),
                };
                let nodes = node
                    .pointer("/projectItems/nodes")
                    .and_then(Value::as_array)
                    .ok_or_else(|| malformed("no list of board items"))?;
                let mut item = None;
                for node in nodes {
                    let on = node
                        .pointer("/project/id")
                        .and_then(Value::as_str)
                        .ok_or_else(|| malformed("a board item naming no board"))?;
                    if on == board {
                        item = Some(node);
                    }
                }
                if let Some(item) = item {
                    let status = match item.get("status") {
                        None | Some(Value::Null) => None,
                        Some(status) => Some(
                            status
                                .get("name")
                                .and_then(Value::as_str)
                                .ok_or_else(|| malformed("a status that names no option"))?
                                .to_owned(),
                        ),
                    };
                    found.insert(
                        id.clone(),
                        (required_nonblank_str(item, "id")?.to_owned(), status),
                    );
                }
            }
        }
        Ok(())
    }

    /// The board item id of `content`, which this process created and remembered without one,
    /// in every half of its own view.
    fn learn_item(&self, content: &NativeId, item: &str) -> Result<(), SourceError> {
        if let Some(held) = self.resolved_cache()?.get_mut(content) {
            held.item_id = item.to_owned();
        }
        if let Some(held) = self.created()?.iter_mut().find(|held| held.id == *content) {
            held.item_id = item.to_owned();
        }
        Ok(())
    }

    /// Ask together which items each of `origins` names, and what creating an item reads; see
    /// [`TaskSource::read_ahead_origins_for_copy`]. An origin whose answer runs past its first
    /// page is left to its own lookup.
    async fn origins_ahead(&self, origins: &[String]) -> Result<(), SourceError> {
        if self.repository.is_none() {
            return Ok(());
        }
        let unasked: Vec<&String> = {
            let held = self.narrowed_cache()?;
            let mut unasked: Vec<&String> = Vec::new();
            for origin in origins {
                if !held.contains_key(&Narrowing::Origin(origin.clone()).key())
                    && !unasked.contains(&origin)
                {
                    unasked.push(origin);
                }
            }
            unasked
        };
        if unasked.is_empty() {
            return Ok(());
        }
        // One batch is held back for the read ahead that follows, which may carry a project's
        // members beside it in the same request.
        if unasked.len() <= ORIGIN_BATCH {
            let mut pending =
                self.pending_origins
                    .lock()
                    .map_err(|_| SourceError::Unavailable {
                        message:
                            "origins held for a read ahead were left inconsistent; run the command \
                          again"
                                .into(),
                    })?;
            for origin in unasked {
                if !pending.contains(origin) {
                    pending.push(origin.clone());
                }
            }
            return Ok(());
        }
        let unasked: Vec<String> = unasked.into_iter().cloned().collect();
        self.send_origin_lookups(&unasked, None).await
    }

    /// Hold, for [`TaskSource::member_classifications`] of `project`, each member's
    /// classification and repositories as the members page `data` carries them — when that page
    /// is every member, each one names `project` as its parent, and every record it holds is
    /// readable. Any other page holds nothing, so the members are read as they always were.
    fn hold_member_classes(&self, project: &NativeId, data: &Value) -> Result<(), SourceError> {
        let Some(page) = data
            .pointer("/members/subIssues")
            .filter(|page| !page.is_null())
        else {
            return Ok(());
        };
        if page
            .pointer("/pageInfo/hasNextPage")
            .and_then(Value::as_bool)
            != Some(false)
        {
            return Ok(());
        }
        let Some(nodes) = page.get("nodes").and_then(Value::as_array) else {
            return Ok(());
        };
        let mut members = Vec::with_capacity(nodes.len());
        for node in nodes {
            if node.pointer("/parent/id").and_then(Value::as_str) != Some(project.0.as_str()) {
                return Ok(());
            }
            let Some(body) = node.get("body").and_then(Value::as_str) else {
                return Ok(());
            };
            // Every node here is filed under `project`, which `BoardKind::of` reads as one of its
            // tasks whatever its marker says, so each is a member the full read would classify.
            let Ok((_, slot)) = metadata_body(Some(body.to_owned())) else {
                return Ok(());
            };
            let Ok(classification) = Classification::from_metadata(&slot) else {
                return Ok(());
            };
            let repositories = if slot.contains_key(Repository::METADATA_KEY) {
                let Ok(repositories) = Repository::from_metadata(&slot) else {
                    return Ok(());
                };
                repositories
            } else {
                let Some(origin) = node
                    .pointer("/repository/nameWithOwner")
                    .and_then(Value::as_str)
                else {
                    return Ok(());
                };
                let Ok(own) = Repository::try_from(format!("{}/{origin}", RepositoryTarget::HOST))
                else {
                    return Ok(());
                };
                vec![own]
            };
            members.push((classification, repositories));
        }
        self.member_classes
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "a project's members held for its classification were left \
                          inconsistent; run the command again"
                    .into(),
            })?
            .insert(project.clone(), members);
        Ok(())
    }

    /// Send every origin held back for the read ahead, beside the members of `members` when it
    /// names a project — or nothing, when none is held.
    async fn send_pending_origins(&self, members: Option<&NativeId>) -> Result<(), SourceError> {
        let pending = std::mem::take(&mut *self.pending_origins.lock().map_err(|_| {
            SourceError::Unavailable {
                message: "origins held for a read ahead were left inconsistent; run the command \
                          again"
                    .into(),
            }
        })?);
        if pending.is_empty() {
            return Ok(());
        }
        self.send_origin_lookups(&pending, members).await
    }

    /// [`graphql::ORIGIN_LOOKUPS`] of `unasked`, [`ORIGIN_BATCH`] to a request — the first one
    /// [`graphql::ORIGIN_LOOKUPS_WITH_MEMBERS`], carrying the members of `members`, when it names
    /// a project — holding each origin's answer and, from a members page that is the whole of
    /// them, each member's classification.
    async fn send_origin_lookups(
        &self,
        unasked: &[String],
        members: Option<&NativeId>,
    ) -> Result<(), SourceError> {
        let Some(repository) = self.repository.clone() else {
            return Ok(());
        };
        let mut members = members;
        for chunk in unasked.chunks(ORIGIN_BATCH) {
            let mut variables = json!({"owner":self.owner,"number":self.project_number,
                "type":"ISSUE","lookupFirst":ORIGIN_LOOKUPS_PAGE_SIZE,"nestedFirst":NESTED_PAGE_SIZE,
                "boardItems":BOARD_ITEMS_PAGE_SIZE,"duplicates":true,
                "repositoryOwner":repository.owner,"repositoryName":repository.name});
            for slot in 0..ORIGIN_BATCH {
                let origin = chunk.get(slot).unwrap_or(&chunk[0]);
                variables[format!("filter{slot}")] =
                    json!(format!("{ORIGIN_FIELD}:{}", quoted(origin)));
                variables[format!("search{slot}")] = json!(
                    self.board_search(Some(&format!("in:body {}", quoted(&as_stored(origin)))))
                );
                variables[format!("ask{slot}")] = json!(slot < chunk.len());
            }
            let project = members.take();
            let document = match project {
                Some(project) => {
                    variables["project"] = json!(project.0);
                    variables["memberFirst"] = json!(MEMBER_PAGE_SIZE);
                    graphql::ORIGIN_LOOKUPS_WITH_MEMBERS
                }
                None => graphql::ORIGIN_LOOKUPS,
            };
            let data = self.graphql(document, variables).await?;
            self.fields_read(&data)?;
            if let Some(project) = project {
                self.hold_member_classes(project, &data)?;
            }
            if let Some(node) = data.get("repository").filter(|value| !value.is_null()) {
                let id = required_nonblank_str(node, "id")?.to_owned();
                self.repository_cache()?.insert(repository.clone(), id);
                if let Some(numeric) = node
                    .get("databaseId")
                    .and_then(Value::as_u64)
                    .and_then(std::num::NonZeroU64::new)
                {
                    self.numeric_repositories
                        .lock()
                        .await
                        .insert(repository.clone(), numeric);
                }
            }
            for (slot, origin) in chunk.iter().enumerate() {
                let items = data
                    .pointer(&format!("/o{slot}/projectV2/items"))
                    .filter(|value| !value.is_null())
                    .ok_or_else(|| SourceError::Refused {
                        message: format!(
                            "GitHub project {}/{} was not found or is not visible to the token",
                            self.owner, self.project_number
                        ),
                    })?;
                let searched =
                    data.get(format!("s{slot}"))
                        .ok_or_else(|| SourceError::Malformed {
                            message: "GitHub search response has no search connection".into(),
                        })?;
                if resumed(items, None)?.has_more() || resumed(searched, None)?.has_more() {
                    continue;
                }
                let mut found: Vec<Resolved> = Vec::new();
                for item in optional_nodes(Some(items), "project items")?
                    .into_iter()
                    .flatten()
                {
                    if let Some(resolved) = self.resolve(item)?
                        && resolved.content_kind == ContentKind::Issue
                        && !found.iter().any(|held| held.id == resolved.id)
                    {
                        found.push(resolved);
                    }
                }
                for node in optional_nodes(Some(searched), "search")?
                    .into_iter()
                    .flatten()
                {
                    if let Some(resolved) = self.resolve_issue(node).await?
                        && !found.iter().any(|held| held.id == resolved.id)
                    {
                        found.push(resolved);
                    }
                }
                self.narrowed_cache()?
                    .insert(Narrowing::Origin((*origin).clone()).key(), found);
            }
        }
        Ok(())
    }

    /// Delete every issue of `ids`, [`DELETE_BATCH`] to a request with
    /// [`graphql::DELETE_ISSUES`], and answer those still there: each a request's answer does not
    /// confirm, every one of a request refused whole included.
    async fn delete_issues(&self, ids: &[&NativeId]) -> Vec<NativeId> {
        let mut left = Vec::new();
        for chunk in ids.chunks(DELETE_BATCH) {
            let mut variables = serde_json::Map::new();
            for slot in 0..DELETE_BATCH {
                let id = chunk.get(slot).unwrap_or(&chunk[0]);
                variables.insert(format!("delete{slot}"), json!({"issueId": id.0}));
                variables.insert(format!("send{slot}"), json!(slot < chunk.len()));
            }
            let data = self
                .graphql(graphql::DELETE_ISSUES, Value::Object(variables))
                .await
                .unwrap_or(Value::Null);
            for (slot, id) in chunk.iter().enumerate() {
                // Confirmed only by the repository the deleted issue was in, named by a
                // non-blank id: anything else says nothing of whether the issue is gone.
                let confirmed = data
                    .pointer(&format!("/delete{slot}/repository/id"))
                    .and_then(Value::as_str)
                    .is_some_and(|id| !id.trim().is_empty());
                if !confirmed {
                    left.push((*id).clone());
                }
            }
        }
        left
    }

    /// Delete one issue, which takes its board item with it.
    async fn delete_issue(&self, id: &NativeId) -> Result<(), SourceError> {
        let data = self
            .graphql(graphql::DELETE_ISSUE, json!({"input":{"issueId":id.0}}))
            .await?;
        data.pointer("/deleteIssue/repository")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub issue deletion returned no repository".into(),
            })?;
        self.forget(id)?;
        Ok(())
    }

    /// Remove one item this copy created, so a copy that could not finish leaves the board
    /// as it found it.
    ///
    /// Deleting the issue takes its board item with it, so there is no second mutation to
    /// keep in step. An id the board does not hold is not an error: the item is already
    /// gone, which is the state this asks for. Which that is, is decided by reading the item
    /// by its own id — a listing of the board can still be missing an item it holds, and
    /// reading that as *already gone* would leave behind the very item this was asked to
    /// take back.
    async fn delete_item(&self, id: &NativeId) -> Result<(), SourceError> {
        let Some(item) = self.bound_item(id).await? else {
            return Ok(());
        };
        if item.content_kind == ContentKind::DraftIssue {
            return Err(SourceError::Refused {
                message: format!(
                    "GitHub item {} is a draft, and this source removes an item by deleting \
                     its issue; next: remove it from the board by hand",
                    id.0
                ),
            });
        }
        let data = self
            .graphql(graphql::DELETE_ISSUE, json!({"input":{"issueId":id.0}}))
            .await?;
        data.pointer("/deleteIssue/repository")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub issue deletion returned no repository".into(),
            })?;
        self.forget(id)?;
        Ok(())
    }

    /// The issue a comment call on `task` is about, or `None` when this board holds no such
    /// task.
    ///
    /// Resolved exactly as [`TaskSource::get_task`] resolves it, so the comment verbs and a
    /// read of the task cannot disagree about which ids name one: a project or a document of
    /// this board is not a task here either.
    ///
    /// A **draft** is a task with nowhere to keep a comment, because GitHub keeps comments on
    /// issues and a draft is not one. It is refused rather than answered with an empty page,
    /// which would read as a task nobody has commented on yet.
    async fn commented_issue(&self, task: &NativeId) -> Result<Option<NativeId>, SourceError> {
        let cached = self.resolved_cache()?.get(task).cloned();
        let Some(item) = (match cached {
            Some(item) => Some(item),
            None => self.item_by_id(task).await?,
        })
        .filter(|item| item.kind == BoardKind::Work(ItemKind::Task)) else {
            return Ok(None);
        };
        if item.content_kind == ContentKind::DraftIssue {
            return Err(self.draft_has_no_comments(task));
        }
        Ok(Some(item.id))
    }

    /// The refusal a comment call on a board draft is answered with: GitHub keeps comments on
    /// issues, and a draft is not one.
    fn draft_has_no_comments(&self, task: &NativeId) -> SourceError {
        SourceError::Refused {
            message: format!(
                "task {} of source {} is a draft item on the board, and GitHub keeps \
                 comments on issues alone, so a draft has none to read or write; next: \
                 convert the draft to an issue on the board, then comment on the issue it \
                 becomes",
                task.0, self.name
            ),
        }
    }

    /// One task and a page of its comments, read with [`graphql::ISSUE_DETAIL`] in one
    /// request — or `None` when this board holds no task by that id.
    ///
    /// What `task show` and a comment listing read. A draft is a task with no comments, so it
    /// is answered with the draft and the refusal, at the price of the draft's own read.
    async fn issue_detail(
        &self,
        id: &NativeId,
        page: &PageRequest,
    ) -> Result<Option<TaskDetailRead>, SourceError> {
        let after = page.cursor.as_ref().map(|cursor| cursor.0.as_str());
        let asked = self
            .graphql(
                graphql::ISSUE_DETAIL,
                json!({"id":id.0,"first":page.limit.min(MAX_PAGE_SIZE),"after":after,
                       "nestedFirst":NESTED_PAGE_SIZE,"boardItems":BOARD_ITEMS_PAGE_SIZE,
                       "duplicates":true}),
            )
            .await;
        let data = match asked {
            Ok(data) => data,
            Err(error) if unresolvable_node(&error) => return Ok(None),
            Err(error) => return Err(error),
        };
        // `node` is null for an id that names nothing, and absent only from an answer this
        // source cannot read — never the same thing.
        let node = data.get("node").ok_or_else(|| SourceError::Malformed {
            message: format!("GitHub answered the read of {} with no node", id.0),
        })?;
        self.detail_of(id, node, true, after).await
    }

    /// Several tasks, each with the first page of its comments when `comments` is set, read
    /// [`DETAIL_BATCH`] at a time with [`graphql::ISSUE_DETAILS`] — one answer per id, in
    /// order.
    ///
    /// A batch GitHub refuses because one of its ids resolves to no node at all is read again
    /// one item at a time, so that id is answered as missing and the others as themselves; any
    /// other refusal is every id of that batch's answer.
    async fn issue_details(
        &self,
        ids: &[NativeId],
        comments: Option<&PageRequest>,
    ) -> Vec<Result<Option<TaskDetailRead>, SourceError>> {
        let mut read = Vec::with_capacity(ids.len());
        for batch in ids.chunks(DETAIL_BATCH) {
            match self
                .graphql(graphql::ISSUE_DETAILS, detail_batch(batch, comments))
                .await
            {
                Ok(data) => {
                    for (slot, id) in batch.iter().enumerate() {
                        // Every alias asked for is answered, null for an id naming nothing;
                        // one missing is an answer this source cannot read.
                        let read_one = match data.get(format!("i{slot}")) {
                            Some(node) => self.detail_of(id, node, comments.is_some(), None).await,
                            None => Err(SourceError::Malformed {
                                message: format!(
                                    "GitHub answered a batch read with no item for {}",
                                    id.0
                                ),
                            }),
                        };
                        read.push(read_one);
                    }
                }
                Err(error) if unresolvable_node(&error) => {
                    for id in batch {
                        read.push(match comments {
                            Some(page) => self.issue_detail(id, page).await,
                            None => self.task_read(id).await,
                        });
                    }
                }
                Err(error) => read.extend(batch.iter().map(|_| Err(error.clone()))),
            }
        }
        read
    }

    /// One task and nothing of its comments, as [`TaskSource::get_task`] reads it.
    async fn task_read(&self, id: &NativeId) -> Result<Option<TaskDetailRead>, SourceError> {
        Ok(self.get_task(id).await?.map(|task| TaskDetailRead {
            task,
            comments: None,
        }))
    }

    /// What one node a detail read reached says: the task this board holds by `id`, with the
    /// page of comments the node carries when `commented` — or `None` for a node that is no
    /// task of this board.
    ///
    /// Resolved as [`Self::item_by_id`] resolves an item: a draft is read again as a draft,
    /// and an item this process created answers from this process's own record, which a node
    /// read taken moments after the write can still be behind.
    async fn detail_of(
        &self,
        id: &NativeId,
        node: &Value,
        commented: bool,
        after: Option<&str>,
    ) -> Result<Option<TaskDetailRead>, SourceError> {
        if node.is_null() {
            return Ok(None);
        }
        let draft = optional_str(node, "__typename")? == Some("DraftIssue");
        // An issue answered under one id is that id's, or the answer is not one this source
        // can report: reporting another issue's task and comments under the qualified id asked
        // for would be the one wrong answer here. A draft's own read checks the same.
        if !draft
            && optional_str(node, "__typename")? == Some("Issue")
            && required_str(node, "id")? != id.0
        {
            return Err(SourceError::Malformed {
                message: format!(
                    "GitHub answered the read of {} with issue {}",
                    id.0,
                    required_str(node, "id")?
                ),
            });
        }
        let item = if draft {
            self.draft_by_id(id).await?
        } else {
            self.resolve_issue(node).await?
        };
        // A read that carried the whole of what the issue is blocked by — a copy's — is kept
        // for this command, so the dependency read the copy makes next is answered from it.
        if let Some(read) = item.as_ref().filter(|item| item.blocked_by.is_some()) {
            self.resolved_cache()?.insert(read.id.clone(), read.clone());
        }
        let Some(item) = item.filter(|item| item.kind == BoardKind::Work(ItemKind::Task)) else {
            return Ok(None);
        };
        let own = self.created()?.iter().find(|own| own.id == *id).cloned();
        let task = own.unwrap_or(item).task()?;
        let comments = match (commented, draft) {
            (false, _) => None,
            (true, true) => Some(Err(self.draft_has_no_comments(id))),
            (true, false) => Some(comment_page(node, &id.0, after).map(Some)),
        };
        if commented && !draft && after.is_none() {
            self.expected.learn_comments(id, node);
        }
        Ok(Some(TaskDetailRead { task, comments }))
    }

    /// Whether the comment `comment` is one of `issue`'s own.
    ///
    /// Read before an edit or a removal is sent, because GitHub's comment mutations take the
    /// comment's id and nothing else: a comment id given against the wrong task would
    /// otherwise change a comment on some other issue entirely. An id that names nothing, or
    /// names something that is not an issue comment, is a comment this task does not have —
    /// which is what GitHub refusing to resolve it means too.
    async fn comment_is_on(
        &self,
        issue: &NativeId,
        comment: &NativeId,
    ) -> Result<bool, SourceError> {
        Ok(self.comment_on(issue, comment).await?.is_some())
    }

    /// The body the comment `comment` holds, exactly as GitHub holds it, when it is one of
    /// `issue`'s own — the read [`Self::comment_is_on`] makes, answering with what it selected.
    async fn comment_on(
        &self,
        issue: &NativeId,
        comment: &NativeId,
    ) -> Result<Option<String>, SourceError> {
        // A write the store's cache expects knows which comments its issue held, and their
        // bodies, at the version it expects: what this read would answer.
        if let Some(primed) = self.expected.primed_comment(issue, comment) {
            return Ok(primed);
        }
        let asked = self
            .graphql(graphql::COMMENT_ISSUE, json!({"id":comment.0}))
            .await;
        let data = match asked {
            Ok(data) => data,
            Err(error) if unresolvable_node(&error) => return Ok(None),
            Err(error) => return Err(error),
        };
        let Some(node) = data.get("node").filter(|value| !value.is_null()) else {
            return Ok(None);
        };
        if optional_str(node, "__typename")? != Some("IssueComment") {
            return Ok(None);
        }
        let on = node.get("issue").ok_or_else(|| SourceError::Malformed {
            message: format!("GitHub issue comment {} names no issue", comment.0),
        })?;
        if required_str(on, "id")? != issue.0 {
            return Ok(None);
        }
        Ok(Some(required_str(node, "body")?.to_owned()))
    }

    /// Which far ends this item's own `blockedBy` relationship holds, and which it cannot.
    async fn partition_edges(
        &self,
        near_kind: BoardKind,
        near_content: ContentKind,
        carried: Option<&[Value]>,
        depends_on: &[DependencyEdge],
    ) -> Result<(Vec<String>, Vec<DependencyEdge>), SourceError> {
        let mut native = Vec::new();
        let mut fallback = Vec::new();
        let far_ends: Vec<(&DependencyEdge, &str, bool, Option<&Value>)> = depends_on
            .iter()
            .map(|edge| {
                // An opaque far end names another tracker, so it is never this board's own.
                let same_source = edge.to.designator().is_none()
                    && edge
                        .to
                        .source()
                        .is_none_or(|source| source == self.name.as_str());
                // A qualified id's source segment runs to its *first* colon — `GlobalId` and
                // `DependencyEndpoint::source` both read it that way — and a native id may hold
                // colons of its own, so the far end is everything after that one separator.
                // Splitting at the last would truncate `work:urn:task:7` to `7`.
                let far_id = if edge.to.is_qualified() {
                    edge.to
                        .id()
                        .split_once(':')
                        .map_or(edge.to.id(), |(_, native)| native)
                } else {
                    edge.to.id()
                };
                // One that already blocks the near issue was answered by that issue's own
                // read, which carried each of its blockers' kinds — an issue every one — so it
                // is not read again.
                let blocking = carried.and_then(|nodes| {
                    nodes
                        .iter()
                        .find(|node| node.get("id").and_then(Value::as_str) == Some(far_id))
                });
                (edge, far_id, same_source, blocking)
            })
            .collect();
        // Every other same-source far end is read by its own id, exactly as the item it is a
        // far end of is: whether this board holds it is that read's answer, never a listing's.
        // They are read together, [`DETAIL_BATCH`] to a request, rather than one each.
        let mut unread: Vec<NativeId> = Vec::new();
        for (_, far_id, same_source, blocking) in &far_ends {
            let id = NativeId((*far_id).to_owned());
            if *same_source && blocking.is_none() && !unread.contains(&id) {
                unread.push(id);
            }
        }
        let read: BTreeMap<NativeId, Option<Resolved>> = unread
            .iter()
            .cloned()
            .zip(self.items_by_ids(&unread).await?)
            .collect();
        for (edge, far_id, same_source, blocking) in far_ends {
            let far = match (same_source, blocking) {
                (false, _) => None,
                (true, Some(node)) => Some(FarEnd {
                    kind: if required_str(node, "title")?.starts_with(DESIGN_TITLE_PREFIX) {
                        BoardKind::Document
                    } else {
                        BoardKind::Work(related_kind(node, self.sub_issues)?)
                    },
                    content_kind: ContentKind::Issue,
                }),
                (true, None) => {
                    let read = read
                        .get(&NativeId(far_id.to_owned()))
                        .cloned()
                        .flatten()
                        .ok_or_else(|| SourceError::Refused {
                            message: format!("GitHub dependency item {far_id} was not found"),
                        })?;
                    Some(FarEnd {
                        kind: read.kind,
                        content_kind: read.content_kind,
                    })
                }
            };
            let far = far.as_ref();
            // The caller says which kind the far end is, and this board holds the far end
            // itself, so a disagreement is settled here rather than stored: recorded, the
            // wrong kind would read back as a cross-level edge that never existed; written
            // natively, it would name a relationship of a different level than the caller
            // asked for.
            //
            // A far end this board holds as a *document* fails the same comparison and is
            // refused by the same sentence: `ItemKind` has no document variant because
            // nothing may point at one, so no caller can name it correctly and the refusal
            // is the only honest answer.
            if let Some(disagreeing) = far.filter(|far| far.kind != BoardKind::Work(edge.to.kind)) {
                return Err(SourceError::Refused {
                    message: format!(
                        "GitHub dependency item {far_id} is a {} of this board, and this item \
                         names it as a {}; record the kind it is",
                        disagreeing.kind.describes(),
                        edge.to.kind.marker()
                    ),
                });
            }
            // A draft has neither `blockedBy` nor `blocking`, so no edge of one is native
            // however the far end is spelled — and one classified native here would be
            // written nowhere at all, because a draft's native reconciliation never runs.
            let native_here = near_content == ContentKind::Issue
                && far.is_some_and(|far| {
                    far.content_kind == ContentKind::Issue
                        && BoardKind::Work(edge.to.kind) == near_kind
                });
            if native_here {
                native.push(far_id.to_owned());
            } else {
                fallback.push(edge.clone());
            }
        }
        Ok((native, fallback))
    }

    async fn update_existing(
        &self,
        item: &Resolved,
        incoming: &Incoming<'_>,
        body: &Option<String>,
        status_target: Option<&StatusTarget>,
    ) -> Result<(), SourceError> {
        let title = incoming.written_title();
        // A terminal status closes the issue here, in the same mutation as its body: its board
        // option was selected before this, so a close never lands on an item whose board cannot
        // show it.
        let mut fields = match item.content_kind {
            ContentKind::DraftIssue => json!({"title":title,"body":body}),
            ContentKind::Issue => json!({"title":title,"body":body,
                                         "stateInput":state_input(status_target)}),
        };
        // The people it is to be assigned to, replacing whoever it holds — sent only when they
        // are not exactly those, and looked up before anything of this write was sent.
        if let Some(people) = incoming.assignees
            && item
                .assignees
                .as_deref()
                .is_none_or(|held| !assignees::same_people(held, people))
        {
            if item.content_kind == ContentKind::DraftIssue {
                return Err(self.assigns_a_draft());
            }
            fields["assigneeIds"] = json!(self.people_ids(people, &item.written_name())?);
        }
        // Only the members whose value differs from the item as this write read it are sent,
        // and nothing at all when none does — whether or not the store's cache expects a
        // version of it. A member left out is one GitHub leaves as it is, so a person's edit
        // of it since that read is kept rather than overwritten with what it already held,
        // and a re-copy of an unchanged item costs no mutation.
        if let Some(members) = fields.as_object_mut() {
            if members.get("title").and_then(Value::as_str) == Some(item.title.as_str()) {
                members.remove("title");
            }
            let body_unchanged = match (members.get("body"), item.raw_body.as_deref()) {
                (Some(Value::String(sent)), Some(held)) => sent == held,
                (Some(Value::Null), held) => held.is_none_or(str::is_empty),
                _ => false,
            };
            if body_unchanged {
                members.remove("body");
            }
            // An issue already in the state the target asks for keeps it: an open one stays
            // open, and a closed one whose closure reads as the target's category keeps its
            // reason — a duplicate's says more than this write would.
            let unchanged_state = match status_target {
                Some(StatusTarget::Terminal(_, reason)) => {
                    item.closed
                        && closed_category(item.reason.as_deref())
                            == closed_category(Some(reason.reason()))
                }
                Some(StatusTarget::Column(_) | StatusTarget::Disabled(_)) => !item.closed,
                None => true,
            };
            if unchanged_state {
                members.remove("stateInput");
            }
            if members.is_empty() {
                return Ok(());
            }
        }
        self.update_content(item.content_kind, &item.id, fields)
            .await
    }

    /// Update one board item's content with exactly `fields` beside its id, through the
    /// mutation its kind takes: `updateIssue` for an issue, `updateProjectV2DraftIssue` for
    /// a draft.
    ///
    /// Every input field either mutation leaves out is a field GitHub leaves as it is, which
    /// is what lets a narrow write carry the one thing it changes and nothing else.
    async fn update_content(
        &self,
        kind: ContentKind,
        id: &NativeId,
        fields: Value,
    ) -> Result<(), SourceError> {
        let (operation, id_key, pointer) = match kind {
            ContentKind::DraftIssue => (
                graphql::UPDATE_DRAFT,
                "draftIssueId",
                "/updateProjectV2DraftIssue/draftIssue",
            ),
            ContentKind::Issue => (graphql::UPDATE_ISSUE, "id", "/updateIssue/issue"),
        };
        if kind == ContentKind::Issue && self.hold_issue(&fields)? {
            return Ok(());
        }
        if kind == ContentKind::Issue && self.hold_update(id, &fields)? {
            return Ok(());
        }
        let mut input = fields;
        input[id_key] = json!(id.0);
        self.send_update(operation, pointer, id, input).await
    }

    /// Send one item's update, `input`, as `operation`, and hold GitHub's answer at `pointer`
    /// to naming that item, `id`.
    async fn send_update(
        &self,
        operation: &str,
        pointer: &str,
        id: &NativeId,
        input: Value,
    ) -> Result<(), SourceError> {
        let data = match self.graphql(operation, json!({"input":&input})).await {
            Ok(data) => data,
            Err(error) => return Err(self.refusing_people(&input, error)),
        };
        let returned = data
            .pointer(pointer)
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub item update returned no item".into(),
            })?;
        if required_str(returned, "id")? != id.0 {
            return Err(SourceError::Malformed {
                message: "GitHub item update returned the wrong item".into(),
            });
        }
        self.assigned_as_asked(&input, Some(returned))
    }

    /// Creates one issue, files it on the board, and reports what a read of it would say:
    /// its content id, its board item id, and the web address GitHub gave it.
    ///
    /// Two calls rather than one: `createIssue` answers with an issue that is on no board,
    /// and `addProjectV2ItemById` is what puts it there. Filing it at creation through
    /// `CreateIssueInput.projectV2Ids` was tried and is not done: GitHub answered with no
    /// board item, and the `addProjectV2ItemById` that then had to follow was refused
    /// "Content already exists in this project". A terminal status is not written here:
    /// `finish_write` selects its option first and closes the issue after, so a close never
    /// lands on an item whose board cannot show it.
    ///
    /// The address and the number come back here because this is the only place either is
    /// known before GitHub's own board read catches up — an item this run created answers
    /// the reads that follow it out of the record below, and one remembered without them
    /// would report no location and no key for the rest of the run.
    async fn create_and_file_issue(
        &self,
        board_id: &str,
        repository: &RepositoryTarget,
        incoming: &Incoming<'_>,
        body: &Option<String>,
    ) -> Result<Landed, SourceError> {
        let repository_id = self.repository_id(repository, incoming).await?;
        let mut input = json!({
            "repositoryId":repository_id,"title":incoming.written_title(),"body":body
        });
        self.assigning(&mut input, incoming)?;
        let data = match self
            .graphql(graphql::CREATE_ISSUE, json!({"input": &input}))
            .await
        {
            Ok(data) => data,
            Err(error) => return Err(self.refusing_people(&input, error)),
        };
        let created = data
            .pointer("/createIssue/issue")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub issue creation returned no issue".into(),
            })?;
        let content_id = NativeId(required_str(created, "id")?.to_owned());
        // Assigned otherwise than asked, the issue is refused and taken back as the filed
        // create's is, a refused delete answered beside the refusal.
        if let Err(error) = self.assigned_as_asked(&input, Some(created)) {
            return Err(self.taken_back(&content_id, created, error).await);
        }
        // Optional although GitHub's schema makes it non-null: the issue exists by now, so
        // a response without it is not worth failing a landed write over — the item simply
        // reports no location until the board read catches up, which is what it did before.
        let url = optional_str(created, "url")?.map(str::to_owned);
        // The issue exists from here on, so an unreadable number and a refused board
        // filing below each try, best effort, to take it back: an issue in the repository
        // that is on no board is an item nobody asked for and nothing here would find again.
        //
        // Its number is optional on the same terms its address is — a landed write is not
        // worth failing over a member that came back missing, and such an item reports no
        // handle until a board read catches up. A number that is *present* and is not an
        // unsigned integer is still a response this source cannot read.
        let number = match created_issue_number(created) {
            Ok(number) => number,
            Err(error) => {
                let _ = self.delete_issue(&content_id).await;
                return Err(error);
            }
        };
        let added = match self
            .graphql(
                graphql::ADD_TO_BOARD,
                json!({"input":{"projectId":board_id,"contentId":content_id.0}}),
            )
            .await
        {
            Ok(added) => added,
            Err(error) => {
                let _ = self.delete_issue(&content_id).await;
                return Err(error);
            }
        };
        let item = added
            .pointer("/addProjectV2ItemById/item")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub board addition returned no project item".into(),
            })?;
        self.note_written(&content_id, url.as_deref());
        Ok(Landed {
            content_id,
            item_id: Some(required_str(item, "id")?.to_owned()),
            url,
            number,
        })
    }

    /// Move one issue under the project it now belongs to, or out of the one it left.
    async fn reparent(
        &self,
        held: Option<NativeId>,
        child: &NativeId,
        wanted: Option<&NativeId>,
    ) -> Result<(), SourceError> {
        if held.as_ref() == wanted {
            return Ok(());
        }
        if let Some(held) = &held {
            self.sub_issue(graphql::REMOVE_SUB_ISSUE, held, child, "removeSubIssue")
                .await?;
        }
        if let Some(wanted) = wanted {
            self.sub_issue(graphql::ADD_SUB_ISSUE, wanted, child, "addSubIssue")
                .await?;
        }
        Ok(())
    }

    async fn sub_issue(
        &self,
        operation: &str,
        parent: &NativeId,
        child: &NativeId,
        root: &str,
    ) -> Result<(), SourceError> {
        let data = self
            .graphql(
                operation,
                json!({"input":{"issueId":parent.0,"subIssueId":child.0}}),
            )
            .await?;
        let issue =
            data.pointer(&format!("/{root}/issue"))
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub sub-issue update returned no issue".into(),
                })?;
        let sub =
            data.pointer(&format!("/{root}/subIssue"))
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub sub-issue update returned no sub-issue".into(),
                })?;
        if required_str(issue, "id")? != parent.0 || required_str(sub, "id")? != child.0 {
            return Err(SourceError::Malformed {
                message: "GitHub sub-issue update returned the wrong issues".into(),
            });
        }
        Ok(())
    }

    /// Bring one issue's `blockedBy` to exactly `native`, sending only the difference, and say
    /// whether there was one.
    ///
    /// An issue [`Issue::Created`] by this very write is blocked by nothing yet, so its
    /// relationships are not read: there is nothing a read of them could find.
    async fn reconcile_blocked_by(
        &self,
        content_id: &NativeId,
        native: &[String],
        issue: Issue<'_>,
    ) -> Result<bool, SourceError> {
        let current = match issue {
            Issue::Created => Vec::new(),
            Issue::Existing(Some(held)) => held
                .iter()
                .map(|far| required_str(far, "id").map(str::to_owned))
                .collect::<Result<Vec<_>, _>>()?,
            Issue::Existing(None) => self.native_dependency_ids(content_id).await?,
        };
        let mut changed = false;
        for (operation, far_id) in current
            .iter()
            .filter(|id| !native.contains(id))
            .map(|id| (graphql::REMOVE_BLOCKED_BY, id))
            .chain(
                native
                    .iter()
                    .filter(|id| !current.contains(id))
                    .map(|id| (graphql::ADD_BLOCKED_BY, id)),
            )
        {
            let data = self
                .graphql(
                    operation,
                    json!({"input":{"issueId":content_id.0,"blockingIssueId":far_id}}),
                )
                .await?;
            let root = if operation == graphql::ADD_BLOCKED_BY {
                "addBlockedBy"
            } else {
                "removeBlockedBy"
            };
            let issue =
                data.pointer(&format!("/{root}/issue"))
                    .ok_or_else(|| SourceError::Malformed {
                        message: "GitHub dependency update returned no issue".into(),
                    })?;
            let blocker = data
                .pointer(&format!("/{root}/blockingIssue"))
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub dependency update returned no blocking issue".into(),
                })?;
            if required_str(issue, "id")? != content_id.0 || required_str(blocker, "id")? != far_id
            {
                return Err(SourceError::Malformed {
                    message: "GitHub dependency update returned the wrong issues".into(),
                });
            }
            changed = true;
        }
        Ok(changed)
    }
}

/// The board field values one write sends, in the order GitHub runs them, and the fields it
/// clears after them: the origin field when it moves, the `Status` option `column` names, the
/// priority, then each projected field.
fn field_writes<'a>(
    existing: Option<&Resolved>,
    origin_field: Option<&str>,
    origin: &str,
    column: Option<(String, String)>,
    priority: Option<&'a PriorityWrite>,
    projections: &'a [ProjectionWrite],
) -> (Vec<(String, Value)>, Vec<&'a str>) {
    let mut fields = Vec::new();
    if let Some(field_id) = origin_field
        && existing.map_or(!origin.is_empty(), |item| {
            item.origin.as_deref().unwrap_or("") != origin
        })
    {
        fields.push((field_id.to_owned(), json!({"text":origin})));
    }
    if let Some((field_id, option_id)) = column {
        fields.push((field_id, json!({"singleSelectOptionId":option_id})));
    }
    let mut clears = Vec::new();
    match priority {
        Some(PriorityWrite::Select { field, option }) => {
            fields.push((field.clone(), json!({"singleSelectOptionId":option})));
        }
        Some(PriorityWrite::Clear { field }) => clears.push(field.as_str()),
        None => {}
    }
    for projection in projections {
        projection.join(&mut fields, &mut clears);
    }
    (fields, clears)
}

/// Everything one write decided before sending anything: see
/// [`GitHubProjectsSource::prepare_write`].
struct Prepared {
    /// What the write lands on.
    destination: WriteDestination,
    board: BoardFields,
    status_target: Option<StatusTarget>,
    /// The `Status` field, the option and the option's name the status lands on.
    // llmlint: ignore[invalid_states_unrepresentable] Exactly what `column_for` returns, which predates this change and four callers share; this record only carries it from `prepare_write` to `complete_write`, which a retyping of that return would reach and this change does not need.
    column: Option<(String, String, String)>,
    priority_write: Option<PriorityWrite>,
    projections: Vec<ProjectionWrite>,
    projected: BTreeMap<String, String>,
    content_kind: ContentKind,
    own_repository: Option<Repository>,
    /// The far ends of the item's native blocked-by relationships.
    // llmlint: ignore[invalid_states_unrepresentable] What `partition_edges` answers and `finish_write` takes as `&[String]`, both of which predate this change; this record only carries the value between them.
    native: Vec<String>,
    slot: BTreeMap<String, Value>,
    content: Option<String>,
    body: Option<String>,
    /// The copy origin the item records, empty for none.
    // llmlint: ignore[invalid_states_unrepresentable] `finish_write` and `field_writes` take the origin as `&str`, empty for none, as they did before this change; this record only carries it to them.
    origin: String,
    origin_field: Option<String>,
}

/// One write a copy held back of the kinds sent together: a board field value, or a blocked-by
/// edge, each its input.
enum HeldWrite<'a> {
    Field(&'a Value),
    Edge(&'a Value),
}

/// Whether `error` is GitHub refusing to file an issue on a board because it is filed there
/// already — what `addProjectV2ItemById` answers once `createIssue`'s `projectV2Ids` has placed
/// it, rather than answering with the item it holds.
fn already_on_the_board(error: &SourceError) -> bool {
    error
        .to_string()
        .contains("Content already exists in this project")
}

/// Hold every alias `inputs` sent under `prefix` in `data` to answering with what its input
/// named — each `(root, named)` of `answers` — and refuse one that answered nothing or another
/// item.
fn check_aliased_answers(
    document: &str,
    data: &Value,
    prefix: &str,
    inputs: &[&Value],
    answers: &[(&str, &str)],
) -> Result<(), SourceError> {
    for (slot, input) in inputs.iter().enumerate() {
        for (root, named) in answers {
            let answered = data
                .pointer(&format!("/{prefix}{slot}/{root}/id"))
                .and_then(Value::as_str)
                .ok_or_else(|| SourceError::Malformed {
                    message: format!(
                        "GitHub answered {} with nothing at {prefix}{slot}",
                        operation_description(document)
                    ),
                })?;
            let asked = input.get(named).and_then(Value::as_str).unwrap_or_default();
            if answered != asked {
                return Err(SourceError::Malformed {
                    message: format!(
                        "GitHub answered {} at {prefix}{slot} with {answered}, which is not \
                         {asked}, the item it was asked to write",
                        operation_description(document)
                    ),
                });
            }
        }
    }
    Ok(())
}

/// `error`, which stopped a batch of creates, and — when GitHub refused to take any of them
/// back — each issue of `left` it created that is still in the repository, which its caller,
/// never handed their ids, could not name.
fn left_behind(error: SourceError, left: &[&str]) -> SourceError {
    if left.is_empty() {
        return error;
    }
    SourceError::Refused {
        message: format!(
            "{error}; GitHub then refused to take back {}, which this write created and which \
             are still in the repository; next: delete {} before writing again",
            left.join(", "),
            if left.len() == 1 { "it" } else { "them" }
        ),
    }
}

/// What one write lands on: the item it updates, as it stood, or where it creates a new one.
enum WriteDestination {
    /// The item as it stood, for an update.
    Existing(Box<Resolved>),
    /// Where a new item is created.
    New(RepositoryTarget),
}

impl WriteDestination {
    /// The item an update writes over, as it stood; `None` for a create.
    fn existing(&self) -> Option<&Resolved> {
        match self {
            Self::Existing(item) => Some(item),
            Self::New(_) => None,
        }
    }
}

/// One existing issue's update held back by [`GitHubProjectsSource::hold_update`]: the issue,
/// and the `UpdateIssueInput` its own [`graphql::UPDATE_ISSUE`] would have sent — the members
/// it writes beside that issue's id, which only [`Self::new`] sets.
struct HeldUpdate {
    issue: NativeId,
    // llmlint: ignore[invalid_states_unrepresentable] The members are exactly what `update_content` would have sent at once, which every one of its callers builds as GraphQL variables; holding them back changes when they are sent, not what they may be, and only `new` builds one, fixing the issue's id. Typing `UpdateIssueInput` would retype every caller of `update_content`, a change of its own.
    input: Value,
}

impl HeldUpdate {
    /// The update writing `members` — an `UpdateIssueInput` without its id — to `issue`.
    fn new(issue: &NativeId, members: &Value) -> Self {
        let mut input = members.clone();
        input["id"] = json!(issue.0);
        Self {
            issue: issue.clone(),
            input,
        }
    }
}

/// What a copy has held back of the items it created; see
/// [`GitHubProjectsSource::hold_back`].
#[derive(Debug, Default)]
enum Deferred {
    /// Writes are being held back, and nothing has been held yet.
    #[default]
    Nothing,
    /// The items held back, every one of them on `board`.
    Held {
        /// The board every one of them is on.
        board: BoardId,
        /// Each created item, in the order it was created.
        created: Vec<HeldBack>,
    },
}

impl Deferred {
    /// Whether `content` is one of the items held back.
    fn holds(&self, content: &NativeId) -> bool {
        match self {
            Self::Nothing => false,
            Self::Held { created, .. } => created.iter().any(|held| held.content == *content),
        }
    }
}

/// One created item's parts that land once [`TaskSource::finish_writes`] is called.
#[derive(Debug)]
struct HeldBack {
    /// The issue.
    content: NativeId,
    /// Its board field values other than its `Status`, in the order they are written.
    fields: Vec<(String, Value)>,
    /// The `Status` field, and the option and its name that the item's status lands on.
    // llmlint: ignore[invalid_states_unrepresentable] The same triple `column_for` returns and `Prepared::column` carries, held back unchanged until it is sent.
    status: Option<(String, String, String)>,
    /// The fields it clears after those are written.
    clears: Vec<String>,
    /// The reason it is closed for, once its option is selected; `None` leaves it open.
    close: Option<ClosedState>,
    /// The issues it is blocked by.
    edges: Vec<NativeId>,
}

/// The member [`answer_with_errors`] carries a response's GraphQL errors under, beside its data:
/// no GitHub type has a field of this name.
const ANSWER_ERRORS: &str = "onetaskgraphErrors";

/// A response's `data` with its GraphQL errors beside it under [`ANSWER_ERRORS`], when it has
/// both — or `None`, for an answer [`GitHubProjectsSource::answer`] reads as it always has.
///
/// What [`graphql::CREATE_ISSUES`] is answered with: GitHub runs a document's mutation fields in
/// order and answers each one that ran, so a refusal of one alias leaves the issues the others
/// created named in the data beside it — the ids a copy needs to take them back.
fn answer_with_errors(body: &str) -> Option<Value> {
    let parsed: Value = serde_json::from_str(body).ok()?;
    let errors = parsed
        .get("errors")?
        .as_array()
        .filter(|errors| !errors.is_empty())?;
    let mut data = parsed.get("data").filter(|data| data.is_object())?.clone();
    data[ANSWER_ERRORS] = Value::Array(errors.clone());
    Some(data)
}

/// The `createIssue` input that files the new issue on `board` and, when `incoming` names one,
/// under its project — see [`GitHubProjectsSource::create_filed_issue`].
fn filed_input(
    repository: &str,
    board: &str,
    incoming: &Incoming<'_>,
    body: &Option<String>,
) -> Value {
    let mut input = json!({"repositoryId":repository,"title":incoming.written_title(),
                           "body":body,"projectV2Ids":[board]});
    if let Some(parent) = incoming.parent {
        input["parentIssueId"] = json!(parent.0);
    }
    input
}

/// The name of option `option` of field `field` among a board's `fields`.
fn option_name<'a>(fields: &'a Value, field: &str, option: &str) -> Option<&'a str> {
    fields
        .get("nodes")?
        .as_array()?
        .iter()
        .find(|node| node.get("id").and_then(Value::as_str) == Some(field))?
        .get("options")?
        .as_array()?
        .iter()
        .find(|candidate| candidate.get("id").and_then(Value::as_str) == Some(option))?
        .get("name")?
        .as_str()
}

/// What a write needs to know of one far end it names: which kind of item it is, and whether
/// it is an issue a native relationship can name.
struct FarEnd {
    kind: BoardKind,
    content_kind: ContentKind,
}

/// Whether the issue one write reconciles was created by that write or was already there.
#[derive(Clone, Copy, PartialEq, Eq)]
enum Issue<'a> {
    /// Created by this write, so it holds no relationships yet.
    Created,
    /// On the board before this write, holding whatever relationships it holds — the far
    /// ends of its whole `blockedBy`, when the read that reached it carried them.
    Existing(Option<&'a [Value]>),
}

/// What resolving one node id reached; see [`GitHubProjectsSource::reach`].
enum Reached {
    /// An issue this board holds, resolved into everything this source reports about it.
    Held(Box<Resolved>),
    /// Nothing this board holds: no such node, or a node on some other board.
    Nothing,
    /// A board draft, which [`graphql::ISSUE`] reaches and reads nothing of, so it is read
    /// again by [`GitHubProjectsSource::draft_by_id`].
    Draft,
}

/// What GitHub says when a string is not a node id it can resolve.
///
/// Matched because it is the ordinary answer to a project selector naming a project by its
/// *name*, and reporting that as a failure would make naming one impossible. It is read
/// off the refusal GitHub sent, never guessed from the shape of the string: this source
/// does not define the syntax of a GitHub node id and would be wrong about it.
const UNRESOLVABLE_NODE: &str = "could not resolve to a node";

/// `error` with `note` added to the end of what it says, its kind and every other member
/// unchanged — so a caller still branches on the failure that happened, and reads beside it
/// what that failure left behind.
fn noting(error: SourceError, note: &str) -> SourceError {
    match error {
        SourceError::Config { message } => SourceError::Config {
            message: message + note,
        },
        SourceError::Auth { message } => SourceError::Auth {
            message: message + note,
        },
        SourceError::Refused { message } => SourceError::Refused {
            message: message + note,
        },
        SourceError::RateLimited {
            retry_after_seconds,
            message,
        } => SourceError::RateLimited {
            retry_after_seconds,
            message: Some(message.unwrap_or_default() + note),
        },
        SourceError::Unavailable { message } => SourceError::Unavailable {
            message: message + note,
        },
        SourceError::Malformed { message } => SourceError::Malformed {
            message: message + note,
        },
    }
}

/// The variables of one [`graphql::ISSUE_COPY_DETAILS`] request, as [`detail_batch`] binds its
/// own: a slot `batch` has no id for is bound to its last id, read again at no added price.
fn copy_detail_batch(batch: &[NativeId], comments: &PageRequest, children: bool) -> Value {
    let mut variables = serde_json::Map::new();
    for slot in 0..COPY_DETAIL_BATCH {
        let id = batch.get(slot).or(batch.last()).map(|id| id.0.clone());
        variables.insert(format!("id{slot}"), json!(id));
    }
    variables.insert("project".to_owned(), json!(batch.first().map(|id| &id.0)));
    variables.insert("children".to_owned(), json!(children));
    variables.insert("first".to_owned(), json!(comments.limit.min(MAX_PAGE_SIZE)));
    variables.insert("nestedFirst".to_owned(), json!(NESTED_PAGE_SIZE));
    variables.insert("boardItems".to_owned(), json!(BOARD_ITEMS_PAGE_SIZE));
    variables.insert("duplicates".to_owned(), json!(true));
    Value::Object(variables)
}

/// The variables of one [`graphql::ISSUE_DETAILS`] request over `batch` — at most
/// [`DETAIL_BATCH`] ids — each item with the first page of its comments when `comments` asks
/// for them.
///
/// The document is fixed-size, so a slot `batch` has no id for is bound to its last id, which
/// is read again at no added price.
fn detail_batch(batch: &[NativeId], comments: Option<&PageRequest>) -> Value {
    let mut variables = serde_json::Map::new();
    for slot in 0..DETAIL_BATCH {
        let id = batch.get(slot).or(batch.last()).map(|id| id.0.clone());
        variables.insert(format!("id{slot}"), json!(id));
    }
    variables.insert(
        "first".to_owned(),
        json!(comments.map_or(MAX_PAGE_SIZE, |page| page.limit.min(MAX_PAGE_SIZE))),
    );
    variables.insert("comments".to_owned(), json!(comments.is_some()));
    variables.insert("nestedFirst".to_owned(), json!(NESTED_PAGE_SIZE));
    variables.insert("boardItems".to_owned(), json!(BOARD_ITEMS_PAGE_SIZE));
    variables.insert("duplicates".to_owned(), json!(true));
    Value::Object(variables)
}

/// Whether this refusal is GitHub saying the id names no node at all.
fn unresolvable_node(error: &SourceError) -> bool {
    matches!(error, SourceError::Refused { message }
        if message.to_ascii_lowercase().contains(UNRESOLVABLE_NODE))
}

/// One project name, as a search qualifier which filters on it at the server.
///
/// Quoted so the whole title is one phrase rather than a bag of words, with the two
/// characters GitHub's own quoting grammar gives a meaning inside a quoted phrase escaped
/// the way it documents. A title matched here is still compared for equality afterwards:
/// the qualifier narrows what the server sends, and this source decides what it names.
fn title_qualifier(name: &str) -> String {
    format!("in:title {}", quoted(name))
}

/// `value` as one quoted phrase of a GitHub search or a board filter, with the two
/// characters GitHub's quoting grammar gives a meaning inside a quoted phrase escaped the way
/// it documents — so a value holding a qualifier's spelling is searched for rather than
/// obeyed.
fn quoted(value: &str) -> String {
    let escaped = value.replace('\\', "\\\\").replace('"', "\\\"");
    format!("\"{escaped}\"")
}

/// The search qualifier for the issues updated at or after `since`.
///
/// Written to the second, rounded down, which can only widen what the search returns.
fn updated_qualifier(since: DateTime<Utc>) -> String {
    format!("updated:>={}", since.format("%Y-%m-%dT%H:%M:%S+00:00"))
}

/// The search terms that narrow a board-scoped issue search to a task query's text and
/// metadata predicates, or `None` when it carries neither.
///
/// The text is one quoted phrase, searched `in:title`, `in:body` or both as its fields say,
/// and each metadata value is one more quoted phrase, which GitHub finds in the body because
/// its index covers the metadata comment the value is stored in. GitHub ANDs the phrases and
/// matches each in any field the `in:` qualifier names, so a query naming a title search and
/// a metadata value searches both fields for both — wider than asked, never narrower, and
/// every candidate is confirmed in process afterwards.
///
/// **This narrows a text search, and that is this source's declared semantics.** GitHub
/// matches whole tokens where a substring rule would match inside a word, so an item holding
/// the text only inside a longer word is not returned. A text of nothing but whitespace
/// matches every item, so it narrows nothing and is not sent.
fn narrowing_qualifiers(query: &TaskQuery) -> Option<String> {
    let text = query
        .text
        .as_ref()
        .filter(|text| !text.terms.trim().is_empty());
    if text.is_none() && query.metadata.is_empty() {
        return None;
    }
    let (title, body) = match text.map(|text| text.fields) {
        None => (false, true),
        Some(TextFields::Title) => (true, !query.metadata.is_empty()),
        Some(TextFields::Content) => (false, true),
        Some(TextFields::TitleOrContent) => (true, true),
    };
    let fields = match (title, body) {
        (true, true) => "in:title,body",
        (true, false) => "in:title",
        _ => "in:body",
    };
    let phrases = text
        .map(|text| text.terms.clone())
        .into_iter()
        .chain(
            query
                .metadata
                .iter()
                .map(|wanted| as_stored(wanted.value())),
        )
        .map(|phrase| quoted(&phrase))
        .collect::<Vec<_>>();
    Some(format!("{fields} {}", phrases.join(" ")))
}

/// Whether every status category `query` keeps reads only off an open issue or a draft: a
/// closed issue reads `done`, `cancelled` or — closed for a reason that implies no category —
/// `unknown`, so a status filter naming none of those three can ask every issue search it sends
/// for open issues alone; each candidate's category is still confirmed from its own item.
fn open_only(query: &TaskQuery) -> bool {
    !query.statuses.is_empty()
        && query.statuses.iter().all(|category| {
            !matches!(
                category,
                StatusCategory::Done | StatusCategory::Cancelled | StatusCategory::Unknown
            )
        })
}

/// What a task query is narrowed by beside a board-scoped issue search's scope — its comment
/// activity and its text and metadata, with `is:open` when [`open_only`] — or `None` for a query
/// carrying none of those, which no narrowed search answers.
///
/// A narrowed search never answers a draft, which is no issue, so narrowing it to open issues
/// too leaves out nothing such a search could have answered.
fn search_qualifiers(query: &TaskQuery) -> Option<String> {
    let mut qualifiers: Vec<String> = [
        query.commented_since.map(updated_qualifier),
        narrowing_qualifiers(query),
    ]
    .into_iter()
    .flatten()
    .collect();
    if qualifiers.is_empty() {
        return None;
    }
    if open_only(query) {
        qualifiers.push(OPEN_QUALIFIER.to_owned());
    }
    Some(qualifiers.join(" "))
}

/// The search terms that narrow a board-scoped issue search to a project or document query's
/// text, or `None` when it has none or a blank one: the phrase, in the fields, a task query
/// carrying that text alone is sent as by [`narrowing_qualifiers`].
fn text_qualifiers(text: Option<&TextQuery>) -> Option<String> {
    narrowing_qualifiers(&TaskQuery {
        text: text.cloned(),
        ..TaskQuery::default()
    })
}

/// Refuses a project or document query's text GitHub's issue search cannot find, before
/// anything is asked of GitHub, on exactly the terms [`refuse_unsearchable`] refuses a task
/// query's.
fn refuse_unsearchable_text(text: Option<&TextQuery>) -> Result<(), SourceError> {
    refuse_unsearchable(&TaskQuery {
        text: text.cloned(),
        ..TaskQuery::default()
    })
}

/// Refuses a task query naming a text or a metadata value GitHub's issue search cannot find,
/// before anything is asked of GitHub.
///
/// GitHub's index holds words, so a phrase with no letter or digit names none to find, and no
/// bounded query answers it: sent, GitHub's answer to it is nothing this source may rely on;
/// left out, the search is every issue of the board. So this source says it cannot answer
/// rather than reading the board or answering nothing. A blank text is not refused: it narrows
/// nothing GitHub could search for, and keeps the board read it always had.
fn refuse_unsearchable(query: &TaskQuery) -> Result<(), SourceError> {
    const WHY: &str = "GitHub's issue search indexes words, so it cannot answer a value with no \
                       letter or digit with a bounded query";
    if let Some(text) = &query.text
        && !text.terms.trim().is_empty()
        && !has_words(&text.terms)
    {
        return Err(SourceError::Refused {
            message: format!(
                "cannot search for the text {:?}: {WHY}; search for a text holding a letter or a digit",
                text.terms
            ),
        });
    }
    if let Some(wanted) = query
        .metadata
        .iter()
        .find(|wanted| !has_words(wanted.value()))
    {
        return Err(SourceError::Refused {
            message: format!(
                "cannot filter by the metadata value {:?} at {:?}: {WHY}; filter by a value holding a letter or a digit",
                wanted.value(),
                std::iter::once(wanted.key())
                    .chain(wanted.path().iter().map(String::as_str))
                    .collect::<Vec<_>>()
                    .join("/"),
            ),
        });
    }
    Ok(())
}

/// Whether GitHub's index could hold a word of `phrase`: whether it has a letter or a digit.
fn has_words(phrase: &str) -> bool {
    phrase.chars().any(char::is_alphanumeric)
}

/// `value` spelled the way the metadata slot stores it: as the inside of its JSON string.
///
/// What GitHub indexes is the slot's JSON text, so a value holding a character JSON escapes —
/// a newline, a tab, a quote — is found by the escape the body holds and not by the character,
/// which GitHub's word match would read as different words.
fn as_stored(value: &str) -> String {
    let encoded = Value::String(value.to_owned()).to_string();
    encoded[1..encoded.len() - 1].to_owned()
}

/// The one narrower question a task query carrying a text, metadata or origin predicate is
/// sent as.
enum Narrowing {
    /// Every carrier of this origin: [`graphql::ORIGIN_LOOKUP`].
    Origin(String),
    /// The board-scoped issue search narrowed by these qualifiers.
    Search(String),
}

impl Narrowing {
    /// What this question is remembered under for the length of one command.
    fn key(&self) -> String {
        match self {
            Self::Origin(origin) => format!("origin {origin}"),
            Self::Search(also) => format!("search {also}"),
        }
    }
}

/// Where one connection of [`graphql::ORIGIN_LOOKUP`] resumes.
enum Resumed {
    /// It reported another page, which starts after this cursor.
    More(String),
    /// It has ended. Sending this cursor again — the page's own end when it had one, and
    /// otherwise the cursor it was reached from — answers an empty page, so the one document
    /// can go on walking the other connection.
    Ended(Option<String>),
}

impl Resumed {
    /// Whether the connection has another page.
    const fn has_more(&self) -> bool {
        matches!(self, Self::More(_))
    }

    /// The cursor to send this connection next.
    fn cursor(self) -> Option<String> {
        match self {
            Self::More(next) => Some(next),
            Self::Ended(last) => last,
        }
    }
}

/// Where `connection`, reached from `after`, resumes — refused when it reports another page
/// with no cursor to it, or from a cursor that does not advance.
fn resumed(connection: &Value, after: Option<&str>) -> Result<Resumed, SourceError> {
    let info = connection
        .get("pageInfo")
        .ok_or_else(|| SourceError::Malformed {
            message: "GitHub connection has no pageInfo".into(),
        })?;
    let end = optional_str(info, "endCursor")?;
    if required_bool(info, "hasNextPage")? {
        let next = end.ok_or_else(|| SourceError::Malformed {
            message: "GitHub connection reports another page and no endCursor".into(),
        })?;
        validate_cursor_progress(after, next)?;
        return Ok(Resumed::More(next.to_owned()));
    }
    Ok(Resumed::Ended(
        end.map(str::to_owned).or_else(|| after.map(str::to_owned)),
    ))
}

/// The board, and every item on it this source reports.
#[derive(Clone)]
struct Board {
    id: String,
    fields: Value,
    items: Vec<Resolved>,
}

/// What a write needs of the board and nothing more: its node id and its field
/// definitions, in the shape a read of the board's own `fields` gives them.
///
/// Deliberately no items. A write decides which item it writes, which parent it files
/// under and which far ends it names by reading each of them by its own id; this is the
/// half of the board those reads cannot carry, and holding no item is what keeps it from
/// ever being asked whether an item is there.
#[derive(Clone, Serialize, Deserialize, schemars::JsonSchema)]
struct BoardFields {
    id: BoardId,
    fields: Value,
}

/// A board's node id: what a field write and `addProjectV2ItemById` address.
///
/// Never blank, because a blank one addresses no board — so an id GitHub answers blank is
/// refused where it is read, and one an item names blank is read as not named at all.
#[derive(Debug, Clone, Serialize, Deserialize, schemars::JsonSchema)]
#[serde(try_from = "String")]
struct BoardId(String);

impl TryFrom<String> for BoardId {
    type Error = String;

    /// The same refusal [`BoardId::parse`] makes, for an id read back from a kept context.
    fn try_from(id: String) -> Result<Self, Self::Error> {
        Self::parse(&id).map_err(|error| error.to_string())
    }
}

/// Where one write left its item, for the record the rest of the command reads it out of.
///
/// A named record rather than a tuple because the update arm and the create arm each fill
/// all four, and two `Option`s of different meaning side by side in a tuple are two
/// positions a reader has to count.
struct Landed {
    /// The issue's own node id, which is the [`NativeId`] this source reports.
    content_id: NativeId,
    /// The board item's id, which is what a field write addresses — `None` for an issue its
    /// create filed on the board itself, whose board item GitHub adds after answering and
    /// [`TaskSource::finish_writes`] reads.
    // llmlint: ignore[invalid_states_unrepresentable] This field and the one below are `Resolved::item_id` and `Resolved::url` carried out of one call: the update arm assigns them from an existing `Resolved` and the whole record is assigned straight back into one. A newtype introduced here alone would be wrapped at both of those boundaries and unwrapped at every use, and would make this private record disagree with the type the same values have on the struct they come from and return to. Where the board item id gets a newtype is on `Resolved`, which is the contract's own shape and not this change's to move.
    item_id: Option<String>,
    /// The web address GitHub gave the issue, when it gave one.
    // llmlint: ignore[invalid_states_unrepresentable] The answer `Resolved::url` and the contract's `Task::url` already record: a web address this source never parses, resolves or compares — it reads GitHub's string and hands it back, and `Location::Url` is where the contract gives it a shape. Validating it here would have this plugin decide what GitHub may call an address.
    url: Option<String>,
    /// The issue's number on its repository, when GitHub reported one.
    number: Option<u64>,
}

impl BoardId {
    fn parse(id: &str) -> Result<Self, SourceError> {
        if id.trim().is_empty() {
            return Err(SourceError::Malformed {
                message: "GitHub named a board with a blank node id".into(),
            });
        }
        Ok(Self(id.to_owned()))
    }

    fn as_str(&self) -> &str {
        &self.0
    }
}

impl Board {
    fn field<'a>(fields: &'a Value, name: &str) -> Result<Option<&'a Value>, SourceError> {
        complete_connection(fields, "project fields", NESTED_PAGE_SIZE)?;
        let nodes = fields
            .get("nodes")
            .and_then(Value::as_array)
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub project fields.nodes is not an array".into(),
            })?;
        Ok(nodes
            .iter()
            .find(|field| field.get("name").and_then(Value::as_str) == Some(name)))
    }
}

/// One board item, resolved into everything this source reports about it.
///
/// Serializable because it is what [`TaskSource::write_context`] hands the store's cache and
/// [`TaskSource::prime_write`] takes back — see [`expected`].
#[derive(Clone, Serialize, Deserialize, schemars::JsonSchema)]
struct Resolved {
    item_id: String,
    id: NativeId,
    content_kind: ContentKind,
    kind: BoardKind,
    title: String,
    body: Option<String>,
    /// The body exactly as GitHub holds it, metadata slot and all, which is what a write
    /// that changes the slot alone has to keep byte for byte outside it.
    raw_body: Option<String>,
    status: Status,
    /// The name of the board `Status` option this item sits in, as the board spells it.
    option: Option<String>,
    /// What its `Priority` field says, read through this instance's mapping.
    priority: HeldPriority,
    /// Whether this item's issue is closed. A draft has no such state and is never closed.
    closed: bool,
    /// GitHub's reason for a closed issue's closure, as it spells it — `None` for an open
    /// issue, a draft, and an issue closed before GitHub recorded reasons.
    #[serde(default)]
    reason: Option<String>,
    /// The tasks this one delivers, read out of its slot. Empty for anything not a task.
    delivers: Vec<TaskRef>,
    /// Every task that delivers this one, read out of its slot. Empty for anything not a
    /// task.
    delivered_by: Vec<TaskRef>,
    labels: Vec<Label>,
    parent: Option<NativeId>,
    /// The issue's GitHub parent on a source reading sub-issues as children (`sub_issues:
    /// parent`), where [`Self::parent`] — the project it is filed under — is always `None`.
    #[serde(default)]
    issue_parent: Option<NativeId>,
    // llmlint: ignore[invalid_states_unrepresentable] The write side's reason, read back: this is the engine's qualified id, taken out of a board text field and handed on untouched. A newtype here would have this plugin define the syntax of an id `docs/metadata.md` says no plugin ever constructs or interprets.
    origin: Option<String>,
    /// What each board text field this instance projects metadata onto holds for this item,
    /// by field name — only the fields holding a value, an empty text included. What a write compares the
    /// item's metadata against, so a field already holding its value is not written again;
    /// never read back as metadata, which the slot alone reports.
    projected: BTreeMap<String, String>,
    /// The issue's own number on its repository, as GitHub reports it.
    ///
    /// `None` in exactly two cases: a draft, which has no number at all — `DraftIssue`
    /// declares none, and a draft is not filed in a repository to be numbered by one — and
    /// an issue this run created whose creating mutation answered without one, which is a
    /// response GitHub's own schema says cannot happen and which a landed write is not
    /// worth failing over. An `Issue` read off the board always has one.
    number: Option<u64>,
    url: Option<String>,
    created_at: Option<DateTime<Utc>>,
    updated_at: Option<DateTime<Utc>>,
    /// When the board item itself last changed — what a board field write moves — when the
    /// read that reached it said.
    item_updated_at: Option<DateTime<Utc>>,
    /// How many comments the issue holds, when the read that reached it said.
    comment_count: Option<u64>,
    /// How many relationship events the issue's timeline holds — every dependency, as
    /// `blockedBy` or `blocking`, and every project membership, as its parent, added or
    /// removed — when the read that reached it said.
    relationship_events: Option<u64>,
    /// The server's clock on the response that read it — its `Date` header — when the
    /// response carried one.
    served_at: Option<DateTime<Utc>>,
    /// The digest of its [`change_token::Facets`] — its own kind, its board field values and
    /// its labels — when the read that reached it carried all three whole.
    #[serde(default)]
    facets: Option<change_token::FacetsDigest>,
    own_repository: Option<Repository>,
    repositories: Vec<Repository>,
    classification: Classification,
    slot: BTreeMap<String, Value>,
    /// The node id of the board this item sits on, when the read that reached it said.
    board_id: Option<String>,
    /// The definition of every board field this item holds a value of, in the shape a read
    /// of the board's own `fields` gives one.
    ///
    /// Only the fields this item has a value in: a field it holds nothing of is not here,
    /// which says nothing about whether the board has it.
    fields: Vec<Value>,
    /// Every field the board this item sits on defines, as its own read of the board's
    /// `fields` gives them — when the read that reached the item carried them, which a read
    /// of it by its own id does. What a write of it needs of the board, then, needs no read
    /// of the board.
    board_fields: Option<Value>,
    /// The far ends of this issue's whole `blockedBy` connection, each as a dependency read
    /// selects one — when the read that reached it carried the connection to its end, which a
    /// read of it by its own id does for any issue blocked by no more than a page. What a
    /// write reconciles that relationship against, and what a read of its forward edges in
    /// the same command answers with.
    blocked_by: Option<Vec<Value>>,
    /// Every person the item is assigned to, by login in the order GitHub answered them — read
    /// only by a source opted in to assignees, and `None` for any other, and for a record no
    /// read or write of this process showed them on.
    #[serde(default)]
    assignees: Option<Vec<String>>,
}

impl Resolved {
    /// The board this item's own read names it on, when that read named one this source can
    /// address.
    fn named_board(&self) -> Option<BoardId> {
        self.board_id
            .as_deref()
            .and_then(|id| BoardId::parse(id).ok())
    }

    /// The board's id and every field it defines, when the read that reached this item
    /// carried both — which a read of it by its own id does.
    fn carried_board(&self) -> Option<BoardFields> {
        Some(BoardFields {
            id: self.named_board()?,
            fields: self.board_fields.clone()?,
        })
    }

    /// Whether this item holds a value of the board field called `name`, and so carries
    /// that field's definition. `false` says nothing about whether the board has the field.
    fn defines(&self, name: &str) -> bool {
        self.fields
            .iter()
            .any(|field| field.get("name").and_then(Value::as_str) == Some(name))
    }

    /// The metadata a caller sees: their own keys, plus the copy origin this source keeps
    /// in a field of its own, and none of the five keys that are only an encoding.
    ///
    /// The two delivery keys are left out for every kind, not only for a task: they are
    /// the encoding of [`Task::delivers`] and [`Task::delivered_by`], and a project or a
    /// document carrying one holds nothing a caller's own metadata could mean by it.
    fn metadata(&self) -> BTreeMap<String, Value> {
        let mut metadata = self.slot.clone();
        metadata.remove(Repository::METADATA_KEY);
        metadata.remove(DependencyEdge::RECORDED_KEY);
        metadata.remove(ItemKind::METADATA_KEY);
        metadata.remove(TaskRef::DELIVERS_KEY);
        metadata.remove(TaskRef::DELIVERED_BY_KEY);
        metadata.remove(Classification::METADATA_KEY);
        // The board field is the origin, and the body's copy of it is only a mirror for the
        // issue search to find: an item whose field holds none has none, whatever its body
        // says, so no reader ever sees two answers.
        metadata.remove(ORIGIN_KEY);
        if let Some(origin) = &self.origin {
            metadata.insert(ORIGIN_KEY.to_owned(), Value::String(origin.clone()));
        }
        metadata
    }

    /// Where this item is, as a link a reader can open.
    ///
    /// A board is a hosted place and every issue on it has a web address, so that address
    /// is what "where is this?" means here — and [`Location::Url`] is what says which kind
    /// of place it is, so a reader knows to open it rather than to read a file out. It
    /// does not replace or derive from `url`: the field goes on reporting exactly what it
    /// reported before, and this says what that address *is*.
    ///
    /// An item GitHub gave no `url` for — a draft has none — reports no location at all
    /// rather than a third variant, which is the contract's "the source did not say". An
    /// issue this run created is not one of those: its address comes back from the
    /// creating mutation, so it is somewhere a reader can open from the moment it exists
    /// rather than from whenever the board read catches up.
    fn location(&self) -> Option<Location> {
        self.url.clone().map(Location::Url)
    }

    /// The short handle this board's backend shows people for a task: the issue's number
    /// alone, as a decimal string.
    ///
    /// The number alone rather than `owner/repo#1043`, because that is the contract's
    /// value for this backend. A draft has no number and so no handle, which is the
    /// contract's *absent* rather than a handle of some other shape — and the native
    /// [`Task::id`] here is the issue's GraphQL node id, which this neither replaces nor
    /// derives from.
    fn key(&self) -> Option<String> {
        self.number.map(|number| number.to_string())
    }

    /// This item's change token — see [`change_token`] — or `None` when the read that
    /// reached it did not carry all six of its parts, which a draft never does, did not say
    /// when GitHub served it, or found it changed too recently to have settled.
    fn version(&self) -> Option<String> {
        let parts = self.token_parts()?;
        change_token::settled(parts.issue, parts.item, self.served_at?)
            .then(|| change_token::spell(&parts))
    }

    /// The change token this item, read just after a write of it, holds now — whether or not it
    /// has settled: what a written task's `version` answers (see
    /// [`GitHubProjectsSource::written_tokens`]), spelled from the parts a later read spells its
    /// token from.
    fn written_version(&self) -> Option<String> {
        self.token_parts().map(|parts| change_token::spell(&parts))
    }

    /// Every part of this item's change token, or `None` when the read that reached it did not
    /// carry them all — a draft never does.
    fn token_parts(&self) -> Option<change_token::Parts> {
        match self.content_kind {
            ContentKind::Issue => Some(change_token::Parts {
                issue: self.updated_at?,
                item: self.item_updated_at?,
                comments: self.comment_count?,
                relationships: self.relationship_events?,
                facets: self.facets.clone()?,
                repository: self.own_repository.clone()?,
            }),
            ContentKind::DraftIssue => None,
        }
    }

    /// Whether its `Priority` field holds a value at all, mapped or not.
    fn holds_priority(&self) -> bool {
        self.priority != HeldPriority::Read(Priority::None)
    }

    /// The task this item is.
    ///
    /// Fails for an item whose `Priority` field holds an option the mapping does not name:
    /// reading that as a level would be a guess, and reading it as `none` would let the next
    /// copy clear a priority a person set.
    fn task(&self) -> Result<Task, SourceError> {
        let priority = match &self.priority {
            HeldPriority::Read(priority) => *priority,
            HeldPriority::Unmapped(option) => {
                return Err(SourceError::Malformed {
                    message: format!(
                        "task {}{} sits in the board {PRIORITY_FIELD} option {option:?}, which \
                         this source's priority_mapping does not name, so its priority cannot be \
                         read; next: name {option:?} under priority_mapping, or move the item to \
                         a mapped option",
                        self.id,
                        self.number
                            .map(|number| format!(" (#{number})"))
                            .unwrap_or_default()
                    ),
                });
            }
        };
        let metadata = self.metadata();
        Ok(Task {
            version: self.version(),
            id: self.id.clone(),
            key: self.key(),
            title: self.title.clone(),
            // The `## Links` section the engine rendered is stripped, so content round-trips.
            content: onetaskgraph_plugin_api::read_content(self.body.clone(), &metadata),
            status: self.status.clone(),
            priority,
            labels: self.labels.clone(),
            project: self.parent.clone(),
            url: self.url.clone(),
            location: self.location(),
            created_at: self.created_at,
            updated_at: self.updated_at,
            metadata,
            repositories: self.repositories.clone(),
            delivers: self.delivers.clone(),
            delivered_by: self.delivered_by.clone(),
            classification: self.classification,
            assignees: self.assignees.clone(),
        })
    }

    fn project(&self) -> Project {
        Project {
            version: self.version(),
            id: self.id.clone(),
            title: self.title.clone(),
            content: self.body.clone(),
            status: self.status.clone(),
            labels: self.labels.clone(),
            url: self.url.clone(),
            location: self.location(),
            created_at: self.created_at,
            updated_at: self.updated_at,
            metadata: self.metadata(),
            repositories: self.repositories.clone(),
            classification: self.classification,
        }
    }

    /// The same issue as a document: the project it is filed under, and no status and no
    /// dependencies, because a document is not work.
    fn document(&self) -> Document {
        Document {
            version: self.version(),
            id: self.id.clone(),
            title: self.title.clone(),
            content: self.body.clone(),
            project: self.parent.clone(),
            labels: self.labels.clone(),
            url: self.url.clone(),
            location: self.location(),
            created_at: self.created_at,
            updated_at: self.updated_at,
            metadata: self.metadata(),
            repositories: self.repositories.clone(),
            classification: self.classification,
        }
    }
}

/// Where one targeted update moves an item's status, and which of its two halves move.
struct StatusMove {
    /// The board the item's `Status` field is on.
    board: BoardId,
    /// The `Status` field's id.
    field: String,
    /// The option's id.
    option: String,
    /// The option's name, as the board spells it.
    name: String,
    /// What the status asks of the issue's state.
    target: StatusTarget,
    /// The status the item reads as once it is there.
    landed: Status,
    /// Which of the status's two halves differ from what the item holds.
    moves: Moves,
}

/// Which halves of an item's status one targeted update moves: its `Status` option, the open or
/// closed state of its issue, or both. A status neither half of which differs is no move at all,
/// and is not a value of this type.
#[derive(Clone, Copy, PartialEq, Eq)]
enum Moves {
    /// The option alone.
    Option,
    /// The issue's state alone: open, closed, or closed with another reason.
    State,
    /// Both.
    Both,
}

impl Moves {
    /// What differs, or `None` when nothing does.
    const fn of(option: bool, state: bool) -> Option<Self> {
        match (option, state) {
            (true, true) => Some(Self::Both),
            (true, false) => Some(Self::Option),
            (false, true) => Some(Self::State),
            (false, false) => None,
        }
    }

    /// Whether the option moves.
    const fn option(self) -> bool {
        matches!(self, Self::Option | Self::Both)
    }

    /// Whether the issue's state moves.
    const fn state(self) -> bool {
        matches!(self, Self::State | Self::Both)
    }
}

/// What one write is, and the status that comes with being it.
///
/// One value rather than a [`BoardKind`] beside an `Option<Status>`: a document has no
/// status and a task or a project always has one, so "a document carrying a status" and
/// "a task carrying none" are states a write cannot be in rather than states every use
/// site below has to defend against.
enum Written<'a> {
    /// A document, which is not work and so has no status at all.
    Document,
    /// A task or a project, and the status it is being written with.
    Work(ItemKind, &'a Status),
}

impl Written<'_> {
    /// Which of the board's three kinds this write is.
    const fn kind(&self) -> BoardKind {
        match self {
            Self::Document => BoardKind::Document,
            Self::Work(kind, _) => BoardKind::Work(*kind),
        }
    }

    /// The status this write carries. A document carries none, so a write of one says
    /// nothing about the issue's open or closed state and selects no board `Status`
    /// option.
    const fn status(&self) -> Option<&Status> {
        match self {
            Self::Document => None,
            Self::Work(_, status) => Some(status),
        }
    }

    /// The status this write carries with the kind whose half of `status_mapping` it is
    /// written through.
    const fn work_status(&self) -> Option<(ItemKind, &Status)> {
        match self {
            Self::Document => None,
            Self::Work(kind, status) => Some((*kind, status)),
        }
    }
}

/// The item being written, in the one shape all three write methods reach.
struct Incoming<'a> {
    written: Written<'a>,
    /// The title a person wrote. A document's goes onto the issue with
    /// [`DESIGN_TITLE_PREFIX`] put back, so a round trip returns the title that went in.
    title: &'a str,
    content: Option<&'a str>,
    assets: Option<&'a onetaskgraph_plugin_api::AssetWrite>,
    labels: &'a [Label],
    metadata: &'a BTreeMap<String, Value>,
    repositories: &'a [Repository],
    /// Recorded in the slot while private, and nowhere while public.
    classification: Classification,
    parent: Option<&'a NativeId>,
    /// [`Task::delivers`], already checked. Empty for a project or a document, which is
    /// what keeps either key out of their slot.
    delivers: &'a [TaskRef],
    /// [`Task::delivered_by`], already checked. Empty for a project or a document.
    delivered_by: &'a [TaskRef],
    /// [`Task::priority`], for a task written to an instance that holds one; `None` for a
    /// project, a document, and every write to an instance with no `priority_mapping` —
    /// which is what keeps such a write's requests exactly what they were before.
    priority: Option<Priority>,
    /// [`Task::assignees`] of a task written to a source opted in to them — the people it is
    /// to be assigned to, replacing whoever it holds — and `None` for everything else, which
    /// leaves an existing item's as they are and creates one unassigned.
    assignees: Option<&'a [String]>,
}

/// What one write does to one projected metadata text field, by the field's id.
#[derive(Debug, Clone)]
enum ProjectionWrite {
    /// Write this text.
    Set {
        field: String, // llmlint: ignore[invalid_states_unrepresentable] A private opaque GraphQL field id passed straight back to GitHub, as `PriorityWrite`'s are; GitHub publishes no grammar for it.
        text: String,
    },
    /// Clear the field's value.
    Clear { field: String }, // llmlint: ignore[invalid_states_unrepresentable] As above.
}

impl ProjectionWrite {
    /// Add this write to a field write's list of values and list of clears.
    fn join<'a>(&'a self, values: &mut Vec<(String, Value)>, clears: &mut Vec<&'a str>) {
        match self {
            Self::Set { field, text } => values.push((field.clone(), json!({"text": text}))),
            Self::Clear { field } => clears.push(field),
        }
    }
}

/// What one write does to an item's `Priority` field.
enum PriorityWrite {
    /// Select this option of this field.
    Select {
        /// The `Priority` field's id.
        field: String,
        /// The mapped option's id.
        option: String,
    },
    /// Clear the field's value, which is what `none` is.
    Clear {
        /// The `Priority` field's id.
        field: String,
    },
}

impl Incoming<'_> {
    /// The title this write puts on the issue.
    fn written_title(&self) -> String {
        match self.written {
            Written::Document => format!("{DESIGN_TITLE_PREFIX}{}", self.title),
            Written::Work(..) => self.title.to_owned(),
        }
    }
}

#[derive(Clone, Copy, PartialEq, Eq, Serialize, Deserialize, schemars::JsonSchema)]
enum ContentKind {
    DraftIssue,
    Issue,
}

/// What one board issue is: a document, or the work an [`ItemKind`] names.
///
/// A type of this source's own rather than an `ItemKind` with a third variant, because
/// `ItemKind` names what a dependency endpoint points at and nothing may point at a
/// document — the contract keeps a document out of that enum deliberately. Holding the
/// board's three answers in one value is what makes every place that asks "which is this?"
/// answer all three, rather than a `document: bool` beside a `kind` that means nothing for
/// two thirds of the board.
#[derive(Clone, Copy, PartialEq, Eq, Serialize, Deserialize, schemars::JsonSchema)]
enum BoardKind {
    /// An issue whose title begins [`DESIGN_TITLE_PREFIX`].
    Document,
    /// Every other issue, and every draft.
    Work(ItemKind),
}

impl BoardKind {
    /// The kind of an item titled `title`, filed under a project or not, holding `sub_issues`
    /// sub-issues and carrying the kind marker `marked` in its slot.
    ///
    /// The design prefix is read *first*, before either of the two rules that separate a
    /// project from a task. A document is not work whatever sub-issues it has and whatever
    /// marker it carries, and reading the prefix later would make a design issue with none of
    /// either an empty project. Being a sub-issue wins outright over the rest, and no marker
    /// overrides it: an issue filed under a project is that project's task even when it has
    /// sub-issues of its own.
    fn of(title: &str, filed: bool, sub_issues: u64, marked: Option<ItemKind>) -> Self {
        if title.starts_with(DESIGN_TITLE_PREFIX) {
            Self::Document
        } else if filed {
            Self::Work(ItemKind::Task)
        } else if sub_issues > 0 || marked == Some(ItemKind::Project) {
            Self::Work(ItemKind::Project)
        } else {
            Self::Work(ItemKind::Task)
        }
    }

    /// What the change token's facets call this kind.
    const fn facet(self) -> change_token::FacetKind {
        match self {
            Self::Document => change_token::FacetKind::Document,
            Self::Work(ItemKind::Project) => change_token::FacetKind::Project,
            Self::Work(ItemKind::Task) => change_token::FacetKind::Task,
        }
    }

    /// Whose half of `status_mapping` an item of this kind reads its status through. A
    /// document has no status of its own, so the task half stands in for whatever the issue
    /// holds; nothing reports it.
    const fn status_kind(self) -> ItemKind {
        match self {
            Self::Document => ItemKind::Task,
            Self::Work(kind) => kind,
        }
    }

    /// How a refusal names this kind to the person reading it.
    const fn describes(self) -> &'static str {
        match self {
            Self::Document => "document",
            Self::Work(kind) => kind.marker(),
        }
    }
}

/// Whether `labels` satisfies `filter`, matching by name, case-insensitively.
///
/// This is the local Markdown source's `labels_match`, spelled the same way on purpose:
/// the shared cross-source journeys assert one answer to one question, so two sources
/// that disagree about what "carries the label bug" means fail them.
fn labels_match(labels: &[Label], filter: &LabelFilter) -> bool {
    let holds = |name: &String| {
        labels
            .iter()
            .any(|label| label.name.eq_ignore_ascii_case(name))
    };
    (filter.any_of.is_empty() || filter.any_of.iter().any(holds))
        && filter.all_of.iter().all(holds)
        && !filter.none_of.iter().any(holds)
}

/// Whether `category` is one of `statuses`. An empty list is unfiltered rather than
/// "keeps nothing", which is what lets a `Vec<StatusCategory>` spell no filter at all.
fn status_matches(category: StatusCategory, statuses: &[StatusCategory]) -> bool {
    statuses.is_empty() || statuses.contains(&category)
}

/// Whether `title`/`content` satisfies `query`, matching case-insensitively.
///
/// `content` is the item's own prose — the body with this source's trailing metadata
/// comment already taken off — so a search never matches an encoding the author of the
/// issue never wrote.
fn text_matches(title: &str, content: Option<&str>, query: &TextQuery) -> bool {
    let terms = query.terms.to_lowercase();
    let in_title = title.to_lowercase().contains(&terms);
    let in_content = content.is_some_and(|body| body.to_lowercase().contains(&terms));
    match query.fields {
        TextFields::Title => in_title,
        TextFields::Content => in_content,
        TextFields::TitleOrContent => in_title || in_content,
    }
}

/// Whether `task` satisfies `query`, with `project` deciding the project predicate.
///
/// The project predicate is passed separately because a read narrowed to one project has
/// already answered it by asking *that project* for its own items — and re-applying it
/// there would compare the caller's selector, which may be a project's **name**, against
/// the id of the project that name resolved to, and keep nothing. Every other read passes
/// `query.project` and applies it here, which is what keeps `projects` a predicate this
/// source really does apply.
fn task_matches(task: &Task, query: &TaskQuery, project: &ProjectFilter) -> bool {
    labels_match(&task.labels, &query.labels)
        && status_matches(task.status.category, &query.statuses)
        && (query.priorities.is_empty() || query.priorities.contains(&task.priority))
        && match project {
            ProjectFilter::Any => true,
            ProjectFilter::Orphans => task.project.is_none(),
            ProjectFilter::Is(id) => task.project.as_ref() == Some(id),
        }
        && query
            .text
            .as_ref()
            .is_none_or(|text| text_matches(&task.title, task.content.as_deref(), text))
        // Against the parsed metadata slot, and against the origin field, which is where
        // `Resolved::metadata` reads each of them from.
        && query.metadata_matches(&task.metadata)
        && query.origin_matches(&task.metadata)
}

fn project_matches(project: &Project, query: &ProjectQuery) -> bool {
    labels_match(&project.labels, &query.labels)
        && status_matches(project.status.category, &query.statuses)
        && query
            .text
            .as_ref()
            .is_none_or(|text| text_matches(&project.title, project.content.as_deref(), text))
}

/// The same three predicates a task query carries, minus the status filter.
///
/// A document is not work, so it has no status for one to compare against and the query
/// type carries none. The project predicate is the same one — a design issue filed under a
/// project issue is in that project, and one filed under nothing is in none — so it is
/// spelled the same way here rather than answered differently.
fn document_matches(document: &Document, query: &DocumentQuery, project: &ProjectFilter) -> bool {
    labels_match(&document.labels, &query.labels)
        && match project {
            ProjectFilter::Any => true,
            ProjectFilter::Orphans => document.project.is_none(),
            ProjectFilter::Is(id) => document.project.as_ref() == Some(id),
        }
        && query
            .text
            .as_ref()
            .is_none_or(|text| text_matches(&document.title, document.content.as_deref(), text))
}

/// Why a GitHub item's change token does not vouch for its forward edges.
///
/// A forward edge names its far end and that far end's kind, and the near item's token moves
/// with the edges it holds but says nothing of the far ends themselves: a far end turning from a
/// task into a project when it gains a sub-issue, retitled into a document, or deleted moves
/// nothing the near token reads, so an edge kept at the near version could name a far end that
/// is no longer what it said.
const FAR_ENDS_UNCOVERED: &str = "a github item's change token does not attest a far end's kind, \
     whether it became a document, or its deletion, so its forward edges are always read from \
     github";

/// The issue-search qualifier that keeps open issues alone, which a status filter naming only
/// categories an open issue reads as narrows a board-scoped search with.
const OPEN_QUALIFIER: &str = "is:open";

/// The board filter `task status reconcile` reads the board through: closed items alone.
const CLOSED_FILTER: &str = "is:closed";

/// The variables of one [`graphql::ISSUES_BY_REFERENCE`] request for `batch`: each slot names
/// one reference, and a slot the batch has none for names its last one again.
fn reference_batch(batch: &[IssueRef]) -> Value {
    let mut variables = serde_json::Map::new();
    for slot in 0..REFERENCE_BATCH {
        let issue = batch.get(slot).or(batch.last());
        variables.insert(format!("owner{slot}"), json!(issue.map(|i| &i.owner)));
        variables.insert(format!("name{slot}"), json!(issue.map(|i| &i.name)));
        variables.insert(format!("number{slot}"), json!(issue.map(|i| i.number)));
    }
    variables.insert("boardItems".to_owned(), json!(BOARD_ITEMS_PAGE_SIZE));
    variables.insert("statusField".to_owned(), json!(STATUS_FIELD));
    variables.insert("duplicates".to_owned(), json!(true));
    Value::Object(variables)
}

/// One issue an issue read by reference answered, in the shape every other read of an issue
/// hands its resolver: each board membership's `Status` value, read by name, as the one field
/// value that membership carries.
///
/// A membership's `status` is `null` when its card sits at no option — or on a board with no
/// field of that name — and otherwise an option value naming its option; anything else is not
/// an answer to this read, and is refused rather than read as a card at no option.
fn held_issue(issue: &Value) -> Result<Value, SourceError> {
    let malformed = |what: &str| SourceError::Malformed {
        message: format!("GitHub answered an issue read by reference with {what}"),
    };
    let mut issue = issue.clone();
    let nodes = issue
        .pointer_mut("/projectItems/nodes")
        .and_then(Value::as_array_mut)
        .ok_or_else(|| malformed("no list of board memberships"))?;
    for node in nodes {
        let node = node
            .as_object_mut()
            .ok_or_else(|| malformed("a board membership that is not an object"))?;
        let status = match node.get("status") {
            None => return Err(malformed("a board membership with no status member")),
            Some(Value::Null) => None,
            Some(status) if status.get("name").is_some_and(Value::is_string) => {
                Some(status.clone())
            }
            // A single-select value of another field type answers `{}`; only an option value
            // carries a name.
            Some(status) if status.as_object().is_some_and(serde_json::Map::is_empty) => None,
            Some(_) => return Err(malformed("a membership status that names no option")),
        };
        node.insert(
            "fieldValues".to_owned(),
            json!({"nodes": status.into_iter().collect::<Vec<_>>(),
                   "pageInfo": {"hasNextPage": false}}),
        );
    }
    Ok(issue)
}

/// The issue one slot of a read by reference answered, `None` when GitHub answered that
/// slot's repository or issue `null` — which [`GitHubProjectsSource::answer_unresolved`]
/// admitted only beside a `NOT_FOUND` naming that very slot — and a refusal when the answer
/// leaves the slot, or its `issue` member, out: an answer missing a member is malformed rather
/// than a reference nothing holds.
fn referenced_issue(data: &Value, slot: usize) -> Result<Option<&Value>, SourceError> {
    let alias = format!("r{slot}");
    let repository = data.get(&alias).ok_or_else(|| SourceError::Malformed {
        message: format!("GitHub answered a read by reference with no {alias}"),
    })?;
    if repository.is_null() {
        return Ok(None);
    }
    let issue = repository
        .get("issue")
        .ok_or_else(|| SourceError::Malformed {
            message: format!("GitHub answered {alias} of a read by reference with no issue"),
        })?;
    if issue.is_null() {
        return Ok(None);
    }
    // `Repository.issue` is typed `Issue`, so anything else is not an answer to this read.
    if optional_str(issue, "__typename")? != Some("Issue") {
        return Err(SourceError::Malformed {
            message: format!(
                "GitHub answered {alias} of a read by reference with an issue whose __typename \
                 is not Issue"
            ),
        });
    }
    Ok(Some(issue))
}

/// A closure reason as a person reads it: `NOT_PLANNED` as `not planned`.
fn spoken_reason(reason: &str) -> String {
    reason.to_ascii_lowercase().replace('_', " ")
}

impl GitHubProjectsSource {
    /// Every task and project of this board whose card contradicts its issue's closure, each
    /// judged against its own kind's half of `status_mapping`, read through the board's own
    /// `is:closed` filter, page by page — never the board's unfiltered `items`.
    ///
    /// Each item the filter answers is confirmed from its own node: an open one, one whose
    /// closure reads `unknown`, and one whose card is already in line are left out whatever
    /// the filter let through. The board's fields ride along and are held for the option
    /// writes that follow, so those need no read of the board.
    async fn contradictions(&self) -> Result<Vec<StatusContradiction>, SourceError> {
        let mut after: Option<String> = None;
        let mut found = Vec::new();
        loop {
            let data = self
                .graphql(
                    graphql::CLOSED_ITEMS,
                    json!({"owner":self.owner,"number":self.project_number,
                           "first":MAX_PAGE_SIZE,"after":after,"filter":CLOSED_FILTER,
                           "nestedFirst":NESTED_PAGE_SIZE,"duplicates":true}),
                )
                .await?;
            let board = data
                .pointer("/closedItems/projectV2")
                .filter(|value| !value.is_null())
                .ok_or_else(|| SourceError::Refused {
                    message: format!(
                        "GitHub project {}/{} was not found or is not visible to the token",
                        self.owner, self.project_number
                    ),
                })?;
            *self.fields_cache()? = Some(BoardFields {
                id: BoardId::parse(required_str(board, "id")?)?,
                fields: board.get("fields").cloned().unwrap_or(Value::Null),
            });
            let items = board.get("items").ok_or_else(|| SourceError::Malformed {
                message: "GitHub project closed items are missing".into(),
            })?;
            for item in items
                .get("nodes")
                .and_then(Value::as_array)
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub project items.nodes is not an array".into(),
                })?
            {
                let Some(resolved) = self.resolve(item)? else {
                    continue;
                };
                let BoardKind::Work(kind) = resolved.kind else {
                    continue;
                };
                let Some((closed_as, name)) = self.statuses.contradicted(
                    kind,
                    resolved.option.as_deref(),
                    resolved.closed,
                    resolved.reason.as_deref(),
                ) else {
                    continue;
                };
                found.push(StatusContradiction {
                    closed_as,
                    to: name.to_owned(),
                    option: resolved.option.clone(),
                    item: match kind {
                        ItemKind::Task => ContradictedItem::Task(resolved.task()?),
                        ItemKind::Project => ContradictedItem::Project(resolved.project()),
                    },
                });
            }
            let info = items
                .get("pageInfo")
                .ok_or_else(|| SourceError::Malformed {
                    message: "GitHub project items have no pageInfo".into(),
                })?;
            if !required_bool(info, "hasNextPage")? {
                return Ok(found);
            }
            let next = required_str(info, "endCursor")?;
            validate_cursor_progress(after.as_deref(), next)?;
            after = Some(next.to_owned());
        }
    }

    /// What this board holds for each issue in `issues`, once each in the order given, read
    /// [`REFERENCE_BATCH`] references at a time — and nothing else of the board.
    ///
    /// An issue GitHub cannot find for this source's credential, and one with no item on this
    /// board, are both absent. A closure that already says the work will not be done here —
    /// not planned, duplicate — is skipped: a landing does not overrule it. An item closed as
    /// completed whose card sits at its kind's `done` option is unchanged, and every other item
    /// is due a `done` write. A project item — a plan's home card, its issue holding the plan's
    /// tasks as sub-issues — is held under exactly these rules, judged against the project
    /// kind's half of `status_mapping`.
    async fn holdings(&self, issues: &[IssueRef]) -> Result<Vec<IssueHolding>, SourceError> {
        let mut distinct: Vec<IssueRef> = Vec::with_capacity(issues.len());
        for issue in issues {
            if !distinct.contains(issue) {
                distinct.push(issue.clone());
            }
        }
        let done = |kind: ItemKind| {
            self.statuses
                .target(kind, StatusCategory::Done)
                .option()
                .unwrap_or("Done")
                .to_owned()
        };
        // GitHub numbers an issue with a GraphQL `Int`, a signed 32-bit value, so a reference past
        // its range names no issue any repository can hold: it is answered absent and never sent,
        // where binding it would have GitHub refuse the whole batch.
        let readable: Vec<IssueRef> = distinct
            .iter()
            .filter(|issue| i32::try_from(issue.number).is_ok())
            .cloned()
            .collect();
        let mut holdings = Vec::with_capacity(readable.len());
        for batch in readable.chunks(REFERENCE_BATCH) {
            let data = self
                .graphql(graphql::ISSUES_BY_REFERENCE, reference_batch(batch))
                .await?;
            for (slot, issue) in batch.iter().enumerate() {
                let node = referenced_issue(&data, slot)?;
                let resolved = match node {
                    Some(node) => self.resolve_issue_as(&held_issue(node)?, false).await?,
                    None => None,
                };
                // A task and a project are both held; a document is not work, so nothing is.
                let Some((item, kind)) = resolved.and_then(|item| match item.kind {
                    BoardKind::Work(kind) => Some((item, kind)),
                    BoardKind::Document => None,
                }) else {
                    holdings.push(IssueHolding {
                        issue: issue.clone(),
                        completion: Completion::Absent,
                    });
                    continue;
                };
                let task = match kind {
                    ItemKind::Task => {
                        let held = item.task()?;
                        HeldTask {
                            id: held.id,
                            kind,
                            option: item.option.clone(),
                            classification: held.classification,
                            repositories: held.repositories,
                            delivers: held.delivers,
                        }
                    }
                    // A project delivers no task, so what its write keeps in step is nothing.
                    ItemKind::Project => {
                        let held = item.project();
                        HeldTask {
                            id: held.id,
                            kind,
                            option: item.option.clone(),
                            classification: held.classification,
                            repositories: held.repositories,
                            delivers: Vec::new(),
                        }
                    }
                };
                let done = done(kind);
                let completion = match (item.closed, item.reason.as_deref()) {
                    (true, Some(reason @ ("NOT_PLANNED" | "DUPLICATE"))) => Completion::Skipped {
                        task,
                        reason: format!("closed as {}", spoken_reason(reason)),
                    },
                    (true, None | Some("COMPLETED"))
                        if item
                            .option
                            .as_deref()
                            .is_some_and(|option| option.eq_ignore_ascii_case(&done)) =>
                    {
                        Completion::Unchanged { task }
                    }
                    _ => {
                        let kept = item.clone();
                        self.held_cache
                            .lock()
                            .map_err(|_| SourceError::Unavailable {
                                message: "items read by issue reference were left \
                                          inconsistent; run the command again"
                                    .into(),
                            })?
                            .insert(kept.id.clone(), kept);
                        Completion::Due {
                            task,
                            to: done.clone(),
                        }
                    }
                };
                holdings.push(IssueHolding {
                    issue: issue.clone(),
                    completion,
                });
            }
        }
        let mut read = holdings.into_iter();
        let mut answered = Vec::with_capacity(distinct.len());
        for issue in distinct {
            if i32::try_from(issue.number).is_ok() {
                answered.push(read.next().ok_or_else(|| SourceError::Malformed {
                    message: format!("no holding was read for {issue}"),
                })?);
            } else {
                answered.push(IssueHolding {
                    issue,
                    completion: Completion::Absent,
                });
            }
        }
        Ok(answered)
    }
}

#[async_trait::async_trait]
impl TaskSource for GitHubProjectsSource {
    fn kind(&self) -> &'static str {
        KIND
    }
    /// GitHub's issue search, which narrows a read of the comments made since an instant, is
    /// an index behind the writes it indexes — the credentialed lane has watched it miss a newly
    /// commented issue for thirty seconds — so a watermark is set back by
    /// [`COMMENT_SEARCH_LAG`], four times that.
    fn comment_consistency_window(&self) -> std::time::Duration {
        COMMENT_SEARCH_LAG
    }
    /// The `Date` of the last answer GitHub gave this source.
    fn served_at(&self) -> Option<DateTime<Utc>> {
        *self
            .last_served
            .lock()
            .unwrap_or_else(std::sync::PoisonError::into_inner)
    }
    fn attach_body_markup(&self, markup: onetaskgraph_plugin_api::SharedBodyMarkup) {
        let _ = self.body_markup.set(markup);
    }
    fn order_versions(&self, a: &str, b: &str) -> Option<std::cmp::Ordering> {
        change_token::order(a, b)
    }
    /// A task's fields and comments: §4.13d of `docs/plugin-protocol.md` states, member by
    /// member, the change each part of the token moves for. Not its forward edges — see
    /// [`FAR_ENDS_UNCOVERED`].
    fn version_coverage(&self) -> VersionCoverage {
        VersionCoverage {
            forward_edges: Coverage::Uncovered {
                reason: FAR_ENDS_UNCOVERED.to_owned(),
            },
            ..VersionCoverage::whole()
        }
    }
    /// Every part, for a copy's skip: the relationship count moves for an edge added or removed
    /// and for a far end deleted, so the token sees every change to a task's edges but one — a
    /// far end's own kind flipping, by gaining a sub-issue or being retitled into a document,
    /// which `spike-coverage` measured stale 1 of 1. A skipping recopy does not carry that, and
    /// the next copy that reads the task does (`docs/plugin-protocol.md` §4.13d).
    // llmlint: ignore[names_match_behavior] The manager's ruling (c-d7838f2c) keeps 0.6.0's GitHub skip and has this plugin state that bounded claim through the coverage it reports, with the bound documented here, in §4.13d and by the copy_skip journey; a separate bounded representation would be a contract change beyond a defaulted addition, which the ruling reserves for the contract's owner, and a follow-up is drafted for a far-end-aware skip.
    fn copy_version_coverage(&self) -> VersionCoverage {
        VersionCoverage::whole()
    }
    /// A project's fields, and not its forward edges, for the same reason a task's are not.
    fn project_version_coverage(&self) -> ProjectVersionCoverage {
        ProjectVersionCoverage {
            fields: Coverage::Covered,
            forward_edges: Coverage::Uncovered {
                reason: FAR_ENDS_UNCOVERED.to_owned(),
            },
        }
    }
    /// A document's fields and the manifest of its image assets, which its metadata slot
    /// records: both are the issue's body, whose every edit moves the token's issue time.
    fn document_version_coverage(&self) -> DocumentVersionCoverage {
        DocumentVersionCoverage::whole()
    }
    fn write_context(&self, id: &NativeId) -> Option<Value> {
        self.context_of(id)
    }
    fn board_context(&self) -> Option<Value> {
        self.board_context_held()
    }
    fn prime_write(&self, id: &NativeId, context: &Value, board: Option<&Value>) -> bool {
        self.prime(id, context, board)
    }
    fn unprime(&self, id: &NativeId) {
        self.unprime_held(id);
    }
    fn placement_context(&self) -> Option<Value> {
        self.placement_held()
    }
    fn prime_placement(&self, context: &Value) -> bool {
        self.prime_placement_held(context)
    }
    fn written_context(&self, id: &NativeId) -> Option<Value> {
        self.written_held(id)
    }
    fn prime_written(&self, id: &NativeId, context: &Value, task: &Task) -> bool {
        self.prime_written_held(id, context, task)
    }
    fn begin_journal(&self) {
        self.expected.begin();
    }
    fn end_journal(&self) -> Option<WriteJournal> {
        Some(self.expected.end())
    }
    fn capabilities(&self) -> Capabilities {
        Capabilities {
            projects: Support::Native,
            documents: Support::Native,
            comments: Support::Native,
            assets: Support::Native,
            priority: if self.priorities.is_some() {
                Support::Native
            } else {
                Support::Unsupported
            },
            filter_by_priority: Support::Native,
            filter_by_comment_activity: Support::Native,
            filter_by_metadata: Support::Native,
            filter_by_origin: Support::Native,
            reconcile_status: Support::Native,
            complete_by_issue: Support::Native,
            orphan_tasks: Support::Native,
            filter_by_label: Support::Native,
            filter_by_status: Support::Native,
            search_title: Support::Native,
            search_content: Support::Native,
            task_dependencies: DependencySupport::BothDirections,
            project_dependencies: DependencySupport::BothDirections,
            max_page_size: MAX_PAGE_SIZE,
            assignees: if self.assignees {
                Support::Native
            } else {
                Support::Unsupported
            },
        }
    }
    async fn visibility(
        &self,
        target: &onetaskgraph_plugin_api::WriteTarget<'_>,
    ) -> Result<onetaskgraph_plugin_api::Visibility, SourceError> {
        self.write_visibility(target).await
    }
    async fn visibility_with_project(
        &self,
        target: &onetaskgraph_plugin_api::WriteTarget<'_>,
        project: &onetaskgraph_plugin_api::WriteTarget<'_>,
    ) -> Result<onetaskgraph_plugin_api::Visibility, SourceError> {
        self.write_visibility_with_project(target, project).await
    }
    fn declare_visibility(&self, declared: onetaskgraph_plugin_api::Visibility) {
        self.declared_private.store(
            declared == onetaskgraph_plugin_api::Visibility::Private,
            std::sync::atomic::Ordering::SeqCst,
        );
    }
    fn scope_visibility<'a>(
        &'a self,
        operation: std::pin::Pin<Box<dyn std::future::Future<Output = ()> + Send + 'a>>,
    ) -> std::pin::Pin<Box<dyn std::future::Future<Output = ()> + Send + 'a>> {
        self.scope_visibility_reads(operation)
    }

    /// Where a create of `target` would file its issue, by [`Self::creation_rule`] — the rule
    /// the create applies — reading the parent project's issue only when `target` names one.
    async fn creation_repository(
        &self,
        target: &onetaskgraph_plugin_api::WriteTarget<'_>,
    ) -> Result<Option<Repository>, SourceError> {
        let onetaskgraph_plugin_api::WriteTarget::New {
            repositories,
            project,
        } = target
        else {
            return Ok(None);
        };
        let target = self
            .creation_rule(repositories, *project, "a new item")
            .await?;
        Repository::try_from(target.origin())
            .map(Some)
            .map_err(|message| SourceError::Malformed { message })
    }
    /// Every issue of the repository a create of `target` files in whose body carries `key` in
    /// its metadata slot, read off that repository's issues newest first back to `since` —
    /// never through search, whose index lags a fresh issue by seconds.
    async fn created_with_key(
        &self,
        record: onetaskgraph_plugin_api::MetadataRecord,
        key: &str,
        since: Option<DateTime<Utc>>,
        target: &onetaskgraph_plugin_api::WriteTarget<'_>,
    ) -> Result<Vec<NativeId>, SourceError> {
        let _ = record;
        let onetaskgraph_plugin_api::WriteTarget::New {
            repositories,
            project,
        } = target
        else {
            return Ok(Vec::new());
        };
        let repository = self
            .creation_rule(repositories, *project, "a new item")
            .await?;
        let wanted = Value::String(key.to_owned());
        let mut found = Vec::new();
        let mut after: Option<String> = None;
        let mut followed = std::collections::HashSet::new();
        loop {
            let data = self
                .graphql(
                    graphql::ISSUES_BY_CREATION,
                    json!({"owner": repository.owner, "name": repository.name,
                           "first": MAX_PAGE_SIZE, "after": after}),
                )
                .await?;
            let issues =
                data.pointer("/repository/issues")
                    .ok_or_else(|| SourceError::Malformed {
                        message: format!(
                            "GitHub answered no issues for repository {}",
                            repository.slug()
                        ),
                    })?;
            let malformed = |what: &str| SourceError::Malformed {
                message: format!(
                    "GitHub answered a page of repository {}'s issues {what}",
                    repository.slug()
                ),
            };
            let nodes = issues
                .get("nodes")
                .and_then(Value::as_array)
                .ok_or_else(|| malformed("with no list of issues"))?;
            let mut older = false;
            for node in nodes {
                let id = node
                    .get("id")
                    .and_then(Value::as_str)
                    .filter(|id| !id.is_empty())
                    .ok_or_else(|| malformed("holding an issue with no id"))?;
                let created = optional_time(node, "createdAt")?
                    .ok_or_else(|| malformed("holding an issue with no creation time"))?;
                if since.is_some_and(|since| created < since) {
                    older = true;
                    break;
                }
                let body = match node.get("body") {
                    Some(Value::String(body)) => Some(body.clone()),
                    Some(Value::Null) => None,
                    _ => return Err(malformed("holding an issue with no body member")),
                };
                let (_, metadata) = metadata_body(body)?;
                if metadata.get(MetadataKey::CREATE_KEY) == Some(&wanted) {
                    found.push(NativeId(id.to_owned()));
                }
            }
            let next = issues
                .pointer("/pageInfo/hasNextPage")
                .and_then(Value::as_bool)
                .ok_or_else(|| malformed("that does not say whether another follows"))?;
            if older || !next {
                return Ok(found);
            }
            // A page that says another follows names where it starts, and never a page already
            // read: anything else would ask for the same pages again for ever.
            let cursor = issues
                .pointer("/pageInfo/endCursor")
                .and_then(Value::as_str)
                .filter(|cursor| !cursor.is_empty())
                .map(str::to_owned);
            if cursor
                .as_ref()
                .is_none_or(|cursor| !followed.insert(cursor.clone()))
            {
                return Err(SourceError::Malformed {
                    message: format!(
                        "GitHub answered a page of repository {}'s issues that says another \
                         follows without a new cursor to it",
                        repository.slug()
                    ),
                });
            }
            after = cursor;
        }
    }

    /// `Repository.issues` lists an issue GitHub committed only once its listing catches up.
    fn create_key_lags(&self) -> bool {
        true
    }

    /// The task as this process's own latest write of it left it, from the item it recorded
    /// that write as — never a request. Its `version` is always `None`: a board's change token
    /// is given only once an item has gone unchanged for [`change_token::SETTLED`], which an
    /// item this command has just written has not, so no write's own answer can vouch for one.
    async fn written_task(&self, id: &NativeId) -> Option<Task> {
        let held = self
            .created()
            .ok()?
            .iter()
            .find(|held| held.id == *id)
            .cloned()
            .or_else(|| {
                self.updated()
                    .ok()?
                    .iter()
                    .find(|held| held.id == *id)
                    .cloned()
            })?;
        let mut task = held.task().ok()?;
        task.version = self.written_token(id);
        Some(task)
    }

    /// Why `creation_owners` refuses a create of `target`, by [`Self::creation_rule`] — read
    /// only when the list is not empty, so a source restricting nothing reads nothing.
    async fn creation_refusal(
        &self,
        target: &onetaskgraph_plugin_api::WriteTarget<'_>,
    ) -> Result<Option<String>, SourceError> {
        let onetaskgraph_plugin_api::WriteTarget::New {
            repositories,
            project,
        } = target
        else {
            return Ok(None);
        };
        if self.creation_owners.is_empty() {
            return Ok(None);
        }
        let target = self
            .creation_rule(repositories, *project, "a new item")
            .await?;
        Ok(self.owner_refusal(&target))
    }

    async fn health(&self) -> Result<Health, SourceError> {
        let board = self.board_page(None, 1).await?;
        Ok(Health {
            reachable: true,
            detail: Some(format!(
                "reading GitHub project {}/{} ({})",
                self.owner,
                self.project_number,
                required_str(&board, "title")?
            )),
        })
    }
    /// The board's own node id, digested: the one request reading its fields, or none when
    /// this command has read them already.
    async fn designator(&self) -> Result<Option<Designator>, SourceError> {
        let board = self.board_fields().await?;
        Designator::of_backend(KIND, board.id.as_str())
            .map(Some)
            .map_err(|message| SourceError::Malformed { message })
    }
    async fn get_task(&self, id: &NativeId) -> Result<Option<Task>, SourceError> {
        self.item_by_id(id)
            .await?
            .filter(|item| item.kind == BoardKind::Work(ItemKind::Task))
            .map(|item| item.task())
            .transpose()
    }
    async fn task_assets(
        &self,
        id: &NativeId,
    ) -> Result<Vec<onetaskgraph_plugin_api::Asset>, SourceError> {
        self.held_assets(id, BoardKind::Work(ItemKind::Task)).await
    }
    async fn task_asset(
        &self,
        id: &NativeId,
        name: &onetaskgraph_plugin_api::AssetName,
    ) -> Result<Option<Vec<u8>>, SourceError> {
        self.held_asset(id, BoardKind::Work(ItemKind::Task), name)
            .await
    }
    async fn set_task_rendering_with_assets(
        &self,
        id: &NativeId,
        content: &str,
        provenance: &Value,
        _answers: &BTreeMap<String, Value>,
        assets: &onetaskgraph_plugin_api::AssetWrite,
    ) -> Result<Option<onetaskgraph_plugin_api::AssetsWritten>, SourceError> {
        self.replace_rendering(
            id,
            BoardKind::Work(ItemKind::Task),
            content,
            provenance,
            Some(assets),
        )
        .await
    }
    async fn document_assets(
        &self,
        id: &NativeId,
    ) -> Result<Vec<onetaskgraph_plugin_api::Asset>, SourceError> {
        self.held_assets(id, BoardKind::Document).await
    }
    async fn document_asset(
        &self,
        id: &NativeId,
        name: &onetaskgraph_plugin_api::AssetName,
    ) -> Result<Option<Vec<u8>>, SourceError> {
        self.held_asset(id, BoardKind::Document, name).await
    }
    async fn set_document_rendering_with_assets(
        &self,
        id: &NativeId,
        content: &str,
        provenance: &Value,
        _answers: &BTreeMap<String, Value>,
        assets: &onetaskgraph_plugin_api::AssetWrite,
    ) -> Result<Option<onetaskgraph_plugin_api::AssetsWritten>, SourceError> {
        self.replace_rendering(id, BoardKind::Document, content, provenance, Some(assets))
            .await
    }
    async fn get_project(&self, id: &NativeId) -> Result<Option<Project>, SourceError> {
        Ok(self
            .item_by_id(id)
            .await?
            .filter(|item| item.kind == BoardKind::Work(ItemKind::Project))
            .map(|item| item.project()))
    }
    /// Every narrowed search and origin lookup asked, a page at a time, as aliased fields of one
    /// [`graphql::SEARCH_BATCH`] request a round — a search's issue search in an `s<n>` field,
    /// an origin's board filter in an `o<n>` field beside its issue search — and every other
    /// query by [`Self::query_tasks`] on its own: one narrowed by a project or by comment
    /// activity, one carrying no narrowing, one resumed from a cursor that read handed out, and
    /// every query of a process that has written this board, whose own writes complete a
    /// narrowed answer. A lone search is its own read too, which costs a batch's point at a
    /// page of its own size.
    ///
    /// Each search's first page asks [`SEARCH_BATCH_FIRST_PAGE`] rows and each later page
    /// [`SEARCH_PAGE_SIZE`]. GitHub answers an error about one aliased field at that field's
    /// path, beside the others' answers, so a refused page fails that query's answer alone; a
    /// request it refuses whole is every query of that request's answer.
    async fn query_tasks_batch(
        &self,
        asks: &[(TaskQuery, PageRequest)],
    ) -> Vec<Result<Page<Task>, SourceError>> {
        self.search_batched(asks).await
    }
    async fn query_tasks(
        &self,
        query: &TaskQuery,
        page: &PageRequest,
    ) -> Result<Page<Task>, SourceError> {
        validate_page(page)?;
        refuse_unsearchable(query)?;
        // A closed issue reads `done`, `cancelled` or — closed for a reason that implies no
        // category — `unknown`; every other category reads only off an open issue or a draft. So
        // a status filter naming none of those three asks every issue search it sends for open
        // issues alone; each candidate's category is still confirmed from its own item below.
        let open_only = open_only(query);
        if query.origin.is_none()
            && !matches!(query.project, ProjectFilter::Is(_))
            && let Some(also) = search_qualifiers(query)
        {
            return self.search_tasks(query, page, &also).await;
        }

        // A read narrowed to one project asks that project for its own tasks, so nothing
        // about it costs what the rest of the board holds. A read carrying a text, metadata
        // or origin predicate asks GitHub the narrower question those predicates are, and a
        // read narrowed to comment activity alone asks the board's own issue search for the
        // issues updated since, which is every issue a comment could have been written or
        // edited on since. Every other task read is a question about the whole board and is
        // answered by reading it.
        let (held, membership) = match (&query.project, query.commented_since) {
            (ProjectFilter::Is(project), _) => (
                match self.held_task_children(project)? {
                    // Taken as `project_children` takes the project's own sub-issue read: with
                    // every item this process wrote under it since, and none it moved away.
                    Some(held) => self.completed_with_written(
                        held.into_iter()
                            .filter(|child| child.parent.as_ref() == Some(project))
                            .collect(),
                        |own| own.parent.as_ref() == Some(project),
                    )?,
                    None => self.project_children(project).await?,
                },
                // Answered by where these items came from; see `task_matches`.
                &ProjectFilter::Any,
            ),
            (ProjectFilter::Any | ProjectFilter::Orphans, since) => {
                match (self.narrowed(query).await?, since) {
                    (Some(narrowed), _) => (narrowed, &query.project),
                    (None, Some(since)) => (self.updated_since(since).await?, &query.project),
                    // The board's own items list every draft; only the search half, which
                    // completes them with issues they lag behind on, is narrowed — a closed
                    // issue that half leaves out is at no category the filter keeps.
                    (None, None) if open_only => (
                        self.board_completed_by(Some(OPEN_QUALIFIER)).await?.items,
                        &query.project,
                    ),
                    (None, None) => (self.board().await?.items, &query.project),
                }
            }
        };
        // Every candidate this read confirms by its comments is read together.
        let candidates = self.comment_candidates(held.iter(), query, membership)?;
        self.hold_comment_pages(&candidates).await?;
        // Filtered before paged: a page of a filtered result is a page of the survivors,
        // never the survivors of a page.
        let mut tasks = Vec::new();
        for item in held
            .iter()
            .filter(|item| item.kind == BoardKind::Work(ItemKind::Task))
        {
            let task = item.task()?;
            if task_matches(&task, query, membership)
                && self.commented_since(item, query.commented_since).await?
            {
                tasks.push(task);
            }
        }
        Ok(offset_page(
            tasks,
            numeric_cursor(page.cursor.as_ref())?,
            page.limit.min(MAX_PAGE_SIZE) as usize,
        ))
    }
    /// Read board pages directly, at the board connection's consistency. Unlike
    /// `query_tasks`, this does not complete a lagging board with issue search: a complete
    /// sweep needs one request per listing page, including the first comments.
    async fn query_task_details(
        &self,
        query: &TaskQuery,
        page: &PageRequest,
        comments: Option<&PageRequest>,
    ) -> Result<Page<TaskDetailRead>, SourceError> {
        let Some(comments) = comments else {
            let tasks = self.query_tasks(query, page).await?;
            return Ok(Page {
                items: tasks
                    .items
                    .into_iter()
                    .map(|task| TaskDetailRead {
                        task,
                        comments: None,
                    })
                    .collect(),
                next: tasks.next,
            });
        };
        validate_page(page)?;
        validate_page(comments)?;
        refuse_unsearchable(query)?;
        if comments.cursor.is_some() || query.commented_since.is_some() {
            return Err(SourceError::Refused {
                message: "a board listing with comments requires a first comment page and cannot filter by comment activity".into(),
            });
        }
        let after = page.cursor.as_ref().map(|cursor| cursor.0.as_str());
        let data = self
            .graphql(
                graphql::BOARD_WITH_COMMENTS,
                json!({
                    "owner":self.owner,"number":self.project_number,
                    "first":page.limit.min(MAX_PAGE_SIZE),"after":after,
                    "commentFirst":comments.limit.min(MAX_PAGE_SIZE),
                    "nestedFirst":NESTED_PAGE_SIZE,"duplicates":true,
                }),
            )
            .await?;
        let connection = data
            .pointer("/owner/projectV2/items")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Refused {
                message: "GitHub project was not found or is not visible to the token".into(),
            })?;
        let mut items = Vec::new();
        for node in optional_nodes(Some(connection), "board items")?
            .into_iter()
            .flatten()
        {
            let Some(item) = self.resolve(node)? else {
                continue;
            };
            if item.kind != BoardKind::Work(ItemKind::Task) {
                continue;
            }
            let task = item.task()?;
            if !task_matches(&task, query, &query.project) {
                continue;
            }
            let read = if item.content_kind == ContentKind::DraftIssue {
                Err(self.draft_has_no_comments(&task.id))
            } else {
                self.expected.learn_comments(&task.id, &node["content"]);
                comment_page(&node["content"], &task.id.0, None).map(Some)
            };
            items.push(TaskDetailRead {
                task,
                comments: Some(read),
            });
        }
        let next = next_cursor(connection)?;
        if let Some(next) = &next {
            validate_cursor_progress(after, &next.0)?;
        }
        Ok(Page { items, next })
    }

    async fn query_projects(
        &self,
        query: &ProjectQuery,
        page: &PageRequest,
    ) -> Result<Page<Project>, SourceError> {
        validate_page(page)?;
        refuse_unsearchable_text(query.text.as_ref())?;
        // The projects a board holds are found by an issue search scoped to that board,
        // never by walking the board's own item connection: what tells a project from a
        // task is the `parent` each issue carries, which costs nothing to read. A query
        // carrying a text asks that search for the text too, so it reads the issues that
        // hold it rather than every issue of the board.
        let held = match self.text_searched(query.text.as_ref()).await? {
            Some(searched) => searched,
            None => self.board_issues().await?,
        };
        let projects = held
            .iter()
            .filter(|item| item.kind == BoardKind::Work(ItemKind::Project))
            .map(Resolved::project)
            .filter(|project| project_matches(project, query))
            .collect();
        Ok(offset_page(
            projects,
            numeric_cursor(page.cursor.as_ref())?,
            page.limit.min(MAX_PAGE_SIZE) as usize,
        ))
    }
    async fn query_projects_by_origin(
        &self,
        origin: &str,
        page: &PageRequest,
    ) -> Result<Page<Project>, SourceError> {
        validate_page(page)?;
        let projects = self
            .copied_from(origin, BoardKind::Work(ItemKind::Project))
            .await?
            .iter()
            .map(Resolved::project)
            .collect();
        Ok(offset_page(
            projects,
            numeric_cursor(page.cursor.as_ref())?,
            page.limit.min(MAX_PAGE_SIZE) as usize,
        ))
    }
    async fn get_document(&self, id: &NativeId) -> Result<Option<Document>, SourceError> {
        let Some(item) = self
            .document_by_id(id)
            .await?
            .filter(|item| item.kind == BoardKind::Document)
        else {
            return Ok(None);
        };
        self.expected.read_document(&item);
        Ok(Some(item.document()))
    }
    async fn query_documents_by_origin(
        &self,
        origin: &str,
        page: &PageRequest,
    ) -> Result<Page<Document>, SourceError> {
        validate_page(page)?;
        let documents = self
            .copied_from(origin, BoardKind::Document)
            .await?
            .iter()
            .map(Resolved::document)
            .collect();
        Ok(offset_page(
            documents,
            numeric_cursor(page.cursor.as_ref())?,
            page.limit.min(MAX_PAGE_SIZE) as usize,
        ))
    }
    async fn query_documents(
        &self,
        query: &DocumentQuery,
        page: &PageRequest,
    ) -> Result<Page<Document>, SourceError> {
        validate_page(page)?;
        // Narrowed to one project, this is the same sub-issue read a task list scoped to
        // that project makes — a document filed under a project is a sub-issue of it too,
        // and which of them come back is the kind this caller asked for. Unscoped, a query
        // carrying a text asks the board-scoped issue search for it, as a task query does,
        // and only one carrying none reads the board.
        let (held, membership) = match &query.project {
            ProjectFilter::Is(project) => (
                self.project_children(project).await?,
                // Answered by where these items came from; see `task_matches`.
                &ProjectFilter::Any,
            ),
            ProjectFilter::Any | ProjectFilter::Orphans => {
                refuse_unsearchable_text(query.text.as_ref())?;
                match self.text_searched(query.text.as_ref()).await? {
                    Some(searched) => (searched, &query.project),
                    None => (self.board().await?.items, &query.project),
                }
            }
        };
        // Filtered before paged, exactly as a task read is: a page of a filtered result is
        // a page of the survivors, never the survivors of a page.
        let documents = held
            .iter()
            .filter(|item| item.kind == BoardKind::Document)
            .map(Resolved::document)
            .filter(|document| document_matches(document, query, membership))
            .collect();
        Ok(offset_page(
            documents,
            numeric_cursor(page.cursor.as_ref())?,
            page.limit.min(MAX_PAGE_SIZE) as usize,
        ))
    }
    async fn labels(&self, page: &PageRequest) -> Result<Page<Label>, SourceError> {
        validate_page(page)?;
        let offset = numeric_cursor(page.cursor.as_ref())?;
        let mut labels = self
            .board()
            .await?
            .items
            .into_iter()
            .flat_map(|item| item.labels)
            .fold(Vec::new(), |mut all, label| {
                if !all.iter().any(|x: &Label| x.id == label.id) {
                    all.push(label);
                }
                all
            });
        labels.sort_by(|a, b| a.name.cmp(&b.name).then(a.id.0.cmp(&b.id.0)));
        Ok(offset_page(
            labels,
            offset,
            page.limit.min(MAX_PAGE_SIZE) as usize,
        ))
    }
    async fn task_dependencies(
        &self,
        id: &NativeId,
        direction: Direction,
        page: &PageRequest,
    ) -> Result<Page<DependencyEdge>, SourceError> {
        self.dependencies(id, ItemKind::Task, direction, page).await
    }
    async fn project_dependencies(
        &self,
        id: &NativeId,
        direction: Direction,
        page: &PageRequest,
    ) -> Result<Page<DependencyEdge>, SourceError> {
        self.dependencies(id, ItemKind::Project, direction, page)
            .await
    }

    fn writes(&self) -> WriteSupport {
        WriteSupport::Supported
    }

    /// Create or update one task.
    ///
    /// Its `delivers` and `delivered_by` are checked before anything is read or written —
    /// neither may name the task itself or name one task twice — and land in the body's
    /// metadata slot under their reserved keys, in place of any caller metadata of those
    /// names.
    async fn write_task(&self, write: &ItemWrite<Task>) -> Result<NativeId, SourceError> {
        if write.item.project.is_some() {
            self.projectless()?;
        }
        self.write_task_assets(write, None)
            .await
            .map(|written| written.id)
    }

    async fn read_ahead_origins_for_copy(&self, origins: &[String]) -> Result<(), SourceError> {
        self.origins_ahead(origins).await
    }

    async fn member_classifications(
        &self,
        project: &NativeId,
    ) -> Result<Option<Vec<(Classification, Vec<Repository>)>>, SourceError> {
        Ok(self
            .member_classes
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "a project's members held for its classification were left \
                          inconsistent; run the command again"
                    .into(),
            })?
            .remove(project))
    }

    /// The issue's own web address, as the `createIssue` that made it in this command answered
    /// it: a task, a project and a document are each an issue here.
    fn written_location(
        &self,
        _record: onetaskgraph_plugin_api::MetadataRecord,
        id: &NativeId,
    ) -> Option<Location> {
        self.written
            .lock()
            .unwrap_or_else(std::sync::PoisonError::into_inner)
            .get(id)
            .cloned()
            .map(Location::Url)
    }

    /// [`MUTATION_BATCH`], while writes are held back: [`graphql::CREATE_ISSUES`] creates that many
    /// in one request, and a copy's tasks are created together only then, since their fields
    /// and edges are what is held back.
    fn create_batch_limit(&self) -> usize {
        if self.deferring().unwrap_or(false) {
            MUTATION_BATCH
        } else {
            1
        }
    }

    // llmlint: ignore[invalid_states_unrepresentable] The signature is the plugin contract's additive `TaskSource::write_tasks`, which the manager approved taking `ItemWrite<Task>` as `write_task` does; a create-only input type would be a new contract type. A write naming an existing target is refused below before anything is sent.
    async fn write_tasks(&self, writes: &[ItemWrite<Task>]) -> Result<Vec<NativeId>, SourceError> {
        if writes.iter().any(|write| write.item.project.is_some()) {
            self.projectless()?;
        }
        if let Some(write) = writes.iter().find(|write| write.target.is_some()) {
            return Err(SourceError::Refused {
                message: format!(
                    "write_tasks creates tasks, and the write of {} names the existing task {}; \
                     next: update it with write_task",
                    write.item.id,
                    write.target.as_ref().map_or("", |target| target.0.as_str())
                ),
            });
        }
        if !self.deferring()? {
            let mut created = Vec::with_capacity(writes.len());
            for write in writes {
                match self.write_task(write).await {
                    Ok(id) => created.push(id),
                    Err(error) => {
                        let mut left: Vec<&str> = Vec::new();
                        for id in &created {
                            if self.delete_issue(id).await.is_err() {
                                left.push(id.0.as_str());
                            }
                        }
                        return Err(left_behind(error, &left));
                    }
                }
            }
            return Ok(created);
        }
        self.create_together(writes).await
    }

    fn begin_deferred_writes(&self) {
        if let Ok(mut deferred) = self.deferred.lock() {
            deferred.get_or_insert_with(Deferred::default);
        }
    }

    async fn finish_writes(&self) -> Result<(), SourceError> {
        self.send_held_updates().await?;
        self.send_held_back().await
    }

    async fn write_task_with_assets(
        &self,
        write: &ItemWrite<Task>,
        _answers: Option<&BTreeMap<String, Value>>,
        assets: &onetaskgraph_plugin_api::AssetWrite,
    ) -> Result<onetaskgraph_plugin_api::AssetsWritten, SourceError> {
        self.write_task_assets(write, Some(assets)).await
    }

    async fn write_project(&self, write: &ItemWrite<Project>) -> Result<NativeId, SourceError> {
        self.projectless()?;
        self.write_item(
            &Incoming {
                written: Written::Work(ItemKind::Project, &write.item.status),
                title: &write.item.title,
                content: write.item.content.as_deref(),
                assets: None,
                labels: &write.item.labels,
                metadata: &write.item.metadata,
                repositories: &write.item.repositories,
                classification: write.item.classification,
                parent: None,
                delivers: &[],
                delivered_by: &[],
                priority: None,
                assignees: None,
            },
            write.target.as_ref(),
            &write.depends_on,
        )
        .await
        .map(|written| written.id)
    }

    fn task_body_size(
        &self,
        write: &ItemWrite<Task>,
        assets: Option<&onetaskgraph_plugin_api::AssetWrite>,
    ) -> Option<onetaskgraph_plugin_api::BodySize> {
        Some(self.body_size(&self.task_incoming(write, assets).ok()?))
    }

    fn project_body_size(
        &self,
        write: &ItemWrite<Project>,
    ) -> Option<onetaskgraph_plugin_api::BodySize> {
        Some(self.body_size(&Incoming {
            written: Written::Work(ItemKind::Project, &write.item.status),
            title: &write.item.title,
            content: write.item.content.as_deref(),
            assets: None,
            labels: &write.item.labels,
            metadata: &write.item.metadata,
            repositories: &write.item.repositories,
            classification: write.item.classification,
            parent: None,
            delivers: &[],
            delivered_by: &[],
            priority: None,
            assignees: None,
        }))
    }

    fn document_body_size(
        &self,
        write: &ItemWrite<Document>,
        assets: Option<&onetaskgraph_plugin_api::AssetWrite>,
    ) -> Option<onetaskgraph_plugin_api::BodySize> {
        Some(self.body_size(&Incoming {
            written: Written::Document,
            title: &write.item.title,
            content: write.item.content.as_deref(),
            assets,
            labels: &write.item.labels,
            metadata: &write.item.metadata,
            repositories: &write.item.repositories,
            classification: write.item.classification,
            parent: write.item.project.as_ref(),
            delivers: &[],
            delivered_by: &[],
            priority: None,
            assignees: None,
        }))
    }

    /// Create or update one document, which is one issue titled the way this board spells
    /// a document.
    ///
    /// Everything else is exactly a task write: caller metadata goes to the same canonical
    /// JSON slot at the end of the body and comes back with its JSON types intact, a key
    /// or a field this board cannot carry is refused by name rather than dropped, a target
    /// naming an issue this board does not hold is refused rather than created, and an
    /// issue this call created is taken back when the rest of the write fails.
    async fn write_document(&self, write: &ItemWrite<Document>) -> Result<NativeId, SourceError> {
        self.write_document_assets(write, None)
            .await
            .map(|written| written.id)
    }

    async fn write_document_with_assets(
        &self,
        write: &ItemWrite<Document>,
        _answers: Option<&BTreeMap<String, Value>>,
        assets: &onetaskgraph_plugin_api::AssetWrite,
    ) -> Result<onetaskgraph_plugin_api::AssetsWritten, SourceError> {
        self.write_document_assets(write, Some(assets)).await
    }

    /// Refused exactly as the write refuses it, from what the write reads: the mapping first,
    /// which reads nothing; then the board's `Status` option. Over an existing item that is
    /// read off the item, as the write reads it, and the item is held among this command's
    /// resolved records so the write that follows reuses that read rather than repeating it;
    /// an item that does not carry the field takes the board's fields, which are held once
    /// read. A create is checked against the board's fields only when this command already
    /// holds them, because a create reads them together with its repository, in one request,
    /// and refuses a missing option before it writes anything.
    async fn check_status_write(
        &self,
        kind: ItemKind,
        category: StatusCategory,
        target: Option<&NativeId>,
    ) -> Result<(), SourceError> {
        let status = self.resolved_target(kind, category)?;
        if status.option().is_none() {
            return Ok(());
        }
        let fields = match target {
            Some(target) => {
                // A target this board does not hold is the write's own refusal to make.
                let Some(item) = self.bound_item(target).await? else {
                    return Ok(());
                };
                self.resolved_cache()?.insert(target.clone(), item.clone());
                // Only the `Status` field decides whether the write can land.
                self.fields_for(Some(&item), false, true, false)
                    .await?
                    .fields
            }
            None => {
                let held = self
                    .board_cache()?
                    .as_ref()
                    .map(|board| board.fields.clone());
                match held.or_else(|| {
                    self.fields_cache()
                        .ok()
                        .and_then(|cache| cache.as_ref().map(|board| board.fields.clone()))
                }) {
                    Some(fields) => fields,
                    None => return Ok(()),
                }
            }
        };
        self.column_for(&fields, kind, category, &status)
            .map(|_| ())
    }

    /// What a write of `status` reads back as: the `Status` option its category maps to for
    /// `kind`, and for a closing category the closure that option and its fixed reason read
    /// as — exactly what [`BoardStatuses::status`] reports of the item once written. A category
    /// this instance has no option for is answered as given, since writing it is refused.
    fn status_as_written(&self, kind: ItemKind, status: &Status) -> Status {
        match self.statuses.target(kind, status.category) {
            StatusTarget::Column(name) => {
                self.statuses.status(kind, Some(name.as_str()), false, None)
            }
            StatusTarget::Terminal(name, reason) => {
                self.statuses
                    .status(kind, Some(name.as_str()), true, Some(reason.reason()))
            }
            StatusTarget::Disabled(_) => status.clone(),
        }
    }

    /// Set one task's status alone.
    ///
    /// An open target reopens a closed issue with an `updateIssue` carrying only its
    /// `stateInput`, then selects the board option with `updateProjectV2ItemFieldValue`; a
    /// terminal target selects its mapped option, then closes with its fixed reason. No
    /// request carries a title, a body or a label. The status
    /// answered is what [`BoardStatuses::status`] reads off the state just written, which is
    /// what a re-read reports.
    async fn set_task_status(
        &self,
        id: &NativeId,
        category: StatusCategory,
    ) -> Result<Option<Status>, SourceError> {
        self.set_status(id, ItemKind::Task, category).await
    }

    /// Set one task's status and answer the task with it, reading the item once: the read the
    /// task is answered from is handed to the status write, which would otherwise read the
    /// same issue again — two requests on a fresh process, the read and the field write,
    /// where the default's `get_task` before `set_task_status` sent three.
    async fn set_task_status_reading(
        &self,
        id: &NativeId,
        category: StatusCategory,
    ) -> Result<Option<Task>, SourceError> {
        // Refused before anything is read, in the words the write itself refuses it.
        self.resolved_target(ItemKind::Task, category)?;
        let Some(item) = self
            .bound_item(id)
            .await?
            .filter(|item| item.kind == BoardKind::Work(ItemKind::Task))
        else {
            return Ok(None);
        };
        let mut task = item.task()?;
        self.held_cache
            .lock()
            .map_err(|_| SourceError::Unavailable {
                message: "items read by issue reference were left inconsistent; run the \
                          command again"
                    .into(),
            })?
            .insert(id.clone(), item);
        let Some(status) = self.set_status(id, ItemKind::Task, category).await? else {
            return Ok(None);
        };
        task.status = status;
        Ok(Some(task))
    }

    /// Set one project's status alone, by exactly the paired write a task's status takes, for
    /// the project kind: a terminal category selects the project kind's mapped option and then
    /// closes the issue with the category's fixed reason — an issue already closed with a
    /// reason that reads that category is left as it is — and an open category reopens a closed
    /// issue and then selects its option. A board draft is refused a terminal category as a
    /// task's is.
    async fn set_project_status(
        &self,
        id: &NativeId,
        category: StatusCategory,
    ) -> Result<Option<Status>, SourceError> {
        self.projectless()?;
        self.set_status(id, ItemKind::Project, category).await
    }

    /// Every task and project whose card contradicts its issue's closure, read through the
    /// board's own `is:closed` filter; see [`TaskSource::status_contradictions`].
    async fn status_contradictions(&self) -> Result<Vec<StatusContradiction>, SourceError> {
        self.contradictions().await
    }

    /// What this board holds for each issue reference, one request per fifty of them; see
    /// [`TaskSource::issue_holdings`].
    async fn issue_holdings(&self, issues: &[IssueRef]) -> Result<Vec<IssueHolding>, SourceError> {
        self.holdings(issues).await
    }

    /// Set one task's priority alone: one `updateProjectV2ItemFieldValue` selecting the
    /// mapped option of the board's `Priority` field, or one `clearProjectV2ItemFieldValue`
    /// for `none`. Refused by an instance with no `priority_mapping`.
    async fn set_task_priority(
        &self,
        id: &NativeId,
        priority: Priority,
    ) -> Result<Option<Priority>, SourceError> {
        self.set_priority(id, priority).await
    }

    /// Replace one task's content with a single body update that keeps the metadata slot
    /// byte for byte.
    async fn set_task_content(
        &self,
        id: &NativeId,
        content: &str,
    ) -> Result<Option<()>, SourceError> {
        self.replace_content(id, content).await
    }

    /// Replace one task issue's content and its provenance slot entry with a single body
    /// update. The answers are not kept: see `replace_rendering`.
    async fn set_task_rendering(
        &self,
        id: &NativeId,
        content: &str,
        provenance: &Value,
        _answers: &BTreeMap<String, Value>,
    ) -> Result<Option<()>, SourceError> {
        self.replace_rendering(
            id,
            BoardKind::Work(ItemKind::Task),
            content,
            provenance,
            None,
        )
        .await
        .map(|written| written.map(|_| ()))
    }

    /// Replace one design-document issue's content and its provenance slot entry, on exactly
    /// the terms of [`set_task_rendering`](TaskSource::set_task_rendering).
    async fn set_document_rendering(
        &self,
        id: &NativeId,
        content: &str,
        provenance: &Value,
        _answers: &BTreeMap<String, Value>,
    ) -> Result<Option<()>, SourceError> {
        self.replace_rendering(id, BoardKind::Document, content, provenance, None)
            .await
            .map(|written| written.map(|_| ()))
    }

    /// Replace one project issue's content and its provenance slot entry, on exactly the
    /// terms of [`set_task_rendering`](TaskSource::set_task_rendering).
    async fn set_project_rendering(
        &self,
        id: &NativeId,
        content: &str,
        provenance: &Value,
        _answers: &BTreeMap<String, Value>,
    ) -> Result<Option<()>, SourceError> {
        self.projectless()?;
        self.replace_rendering(
            id,
            BoardKind::Work(ItemKind::Project),
            content,
            provenance,
            None,
        )
        .await
        .map(|written| written.map(|_| ()))
    }

    /// Apply a targeted update with one read of the item and a write only for what differs:
    /// the `Status` and `Priority` field writes in one request, the `blockedBy` difference,
    /// and last one `updateIssue` for title, body and state. See `targeted_update`.
    async fn update_task(
        &self,
        id: &NativeId,
        update: &TaskUpdate,
    ) -> Result<Option<TaskUpdateOutcome>, SourceError> {
        self.targeted_update(id, update).await
    }

    /// Each issue's sub-issue parent and children and its links, read in batches.
    async fn read_relations(
        &self,
        ids: &[NativeId],
    ) -> Vec<Result<Option<onetaskgraph_plugin_api::Relations>, SourceError>> {
        self.relations_of(ids).await
    }

    async fn set_parents(
        &self,
        writes: &[onetaskgraph_plugin_api::ParentWrite],
    ) -> Vec<Result<onetaskgraph_plugin_api::ParentAnswer, SourceError>> {
        self.parents_set(writes).await
    }

    fn link_types(&self) -> onetaskgraph_plugin_api::LinkTypes {
        self.link_types.clone()
    }

    async fn add_native_links(
        &self,
        links: &[onetaskgraph_plugin_api::NativeLink],
    ) -> Vec<Result<(), SourceError>> {
        self.native_links(links, relations::NativeWrite::Add).await
    }

    async fn remove_native_links(
        &self,
        links: &[onetaskgraph_plugin_api::NativeLink],
    ) -> Vec<Result<(), SourceError>> {
        self.native_links(links, relations::NativeWrite::Remove)
            .await
    }

    /// Every task read together — [`DETAIL_BATCH`] to a [`graphql::ISSUE_DETAILS`] request —
    /// and each update's hold and guard decided on that read; then each update the read lets
    /// through composed exactly as [`Self::update_task`] composes one, its board field writes
    /// and its issue update held back and sent in [`graphql::WRITE_TASKS`] requests: every
    /// field write first, then the issue update of every task whose field writes all landed.
    ///
    /// **Where GitHub would refuse the batch as a whole, each item is read on its own:** a
    /// batched node read naming an id that resolves to nothing is refused whole, so each id of
    /// that batch is read by itself, the missing one answered missing and the rest as
    /// themselves. A batch of one, an update of a board draft, one naming dependencies and one
    /// of an item whose body this process has not read are each made on their own, as
    /// [`Self::update_task`] makes them, because each sends writes of its own in an order the
    /// batch does not hold. A field write GitHub refuses fails that task's update alone and
    /// sends nothing of its body.
    async fn update_tasks(
        &self,
        updates: &[onetaskgraph_plugin_api::BatchUpdate],
    ) -> Vec<Result<onetaskgraph_plugin_api::BatchAnswer<TaskUpdateOutcome>, SourceError>> {
        self.updates_batched(updates).await
    }

    /// Replace one task's `delivered_by` with a single body update that changes the
    /// metadata slot and nothing outside it.
    async fn set_delivered_by(
        &self,
        id: &NativeId,
        delivered_by: &[TaskRef],
    ) -> Result<Option<()>, SourceError> {
        self.replace_delivered_by(id, delivered_by).await
    }

    /// Set one key of one task issue's metadata with a single body update that changes the
    /// metadata slot and nothing outside it — no title, label, state or board field request —
    /// and sends nothing when the task already holds that value under the key.
    async fn set_task_metadata(
        &self,
        id: &NativeId,
        key: &MetadataKey,
        value: &Value,
    ) -> Result<Option<Task>, SourceError> {
        Ok(self
            .set_slot_key(id, BoardKind::Work(ItemKind::Task), key, value)
            .await?
            .map(|item| item.task())
            .transpose()?)
    }

    /// Set one key of one project issue's metadata, on exactly the terms of
    /// [`set_task_metadata`](TaskSource::set_task_metadata).
    async fn set_project_metadata(
        &self,
        id: &NativeId,
        key: &MetadataKey,
        value: &Value,
    ) -> Result<Option<Project>, SourceError> {
        self.projectless()?;
        Ok(self
            .set_slot_key(id, BoardKind::Work(ItemKind::Project), key, value)
            .await?
            .map(|item| item.project()))
    }

    /// Set one key of one design-document issue's metadata, on exactly the terms of
    /// [`set_task_metadata`](TaskSource::set_task_metadata).
    async fn set_document_metadata(
        &self,
        id: &NativeId,
        key: &MetadataKey,
        value: &Value,
    ) -> Result<Option<Document>, SourceError> {
        Ok(self
            .set_slot_key(id, BoardKind::Document, key, value)
            .await?
            .map(|item| item.document()))
    }

    /// Take one key out of one issue's metadata slot with a single body update that changes the
    /// slot and nothing outside it, sending nothing when the slot does not hold the key.
    async fn remove_record_metadata(
        &self,
        record: onetaskgraph_plugin_api::MetadataRecord,
        id: &NativeId,
        key: &MetadataKey,
    ) -> Result<Option<Option<String>>, SourceError> {
        let kind = match record {
            onetaskgraph_plugin_api::MetadataRecord::Task => BoardKind::Work(ItemKind::Task),
            onetaskgraph_plugin_api::MetadataRecord::Project => BoardKind::Work(ItemKind::Project),
            onetaskgraph_plugin_api::MetadataRecord::Document => BoardKind::Document,
        };
        let Some(mut item) = self.bound_item(id).await?.filter(|item| item.kind == kind) else {
            return Ok(None);
        };
        if item.slot.contains_key(key.as_str()) {
            let mut slot = item.slot.clone();
            slot.remove(key.as_str());
            self.write_slot(&mut item, &slot).await?;
            self.remember_written(item.clone(), false)?;
        }
        Ok(Some(match record {
            onetaskgraph_plugin_api::MetadataRecord::Task => item.task()?.version,
            onetaskgraph_plugin_api::MetadataRecord::Project => item.project().version,
            onetaskgraph_plugin_api::MetadataRecord::Document => item.document().version,
        }))
    }

    async fn delete_task(&self, id: &NativeId) -> Result<(), SourceError> {
        self.delete_item(id).await
    }

    async fn delete_project(&self, id: &NativeId) -> Result<(), SourceError> {
        self.projectless()?;
        self.delete_item(id).await
    }

    async fn delete_document(&self, id: &NativeId) -> Result<(), SourceError> {
        self.delete_item(id).await
    }

    /// One page of the task issue's own comments, walked by GitHub's own cursor.
    ///
    /// Nothing here filters, so nothing has to be read ahead of the page: the caller's limit is
    /// the page GitHub is asked for and GitHub's `endCursor` is the cursor handed back.
    ///
    /// One request, [`graphql::ISSUE_DETAIL`]: the read that says the id names a task of this
    /// board is the read of its comments. A draft this process already resolved is refused
    /// without one.
    async fn task_comments(
        &self,
        task: &NativeId,
        page: &PageRequest,
    ) -> Result<Option<Page<Comment>>, SourceError> {
        validate_page(page)?;
        let cached = self.resolved_cache()?.get(task).cloned();
        if let Some(item) = cached {
            if item.kind != BoardKind::Work(ItemKind::Task) {
                return Ok(None);
            }
            if item.content_kind == ContentKind::DraftIssue {
                return Err(self.draft_has_no_comments(task));
            }
        }
        // The first page a comment-activity read just read of this task, at the size asked, is
        // that read's answer: taken, so it answers once.
        if page.cursor.is_none()
            && page.limit >= MAX_PAGE_SIZE
            && let Some(held) = self.held_comments()?.remove(task)
        {
            return Ok(Some(held));
        }
        match self.issue_detail(task, page).await? {
            Some(TaskDetailRead {
                comments: Some(comments),
                ..
            }) => comments,
            _ => Ok(None),
        }
    }

    /// Every id's task, with the first page of its comments when `comments` names it:
    /// [`DETAIL_BATCH`] items per [`graphql::ISSUE_DETAILS`] request, and one item with its
    /// comments in one [`graphql::ISSUE_DETAIL`] request.
    async fn get_task_details(
        &self,
        ids: &[NativeId],
        comments: Option<&PageRequest>,
    ) -> Vec<Result<Option<TaskDetailRead>, SourceError>> {
        if let Some(page) = comments
            && let Err(error) = validate_page(page)
        {
            return ids.iter().map(|_| Err(error.clone())).collect();
        }
        match (ids, comments) {
            ([id], Some(page)) => vec![self.issue_detail(id, page).await],
            ([id], None) => vec![self.task_read(id).await],
            _ => self.issue_details(ids, comments).await,
        }
    }

    /// Every id's task with the first page of its comments and its `blockedBy`, as a copy reads
    /// the tasks it carries: [`COPY_DETAIL_BATCH`] items per [`graphql::ISSUE_COPY_DETAILS`]
    /// request, so the dependency read a copy makes of each next is answered from this one.
    async fn get_task_details_for_copy(
        &self,
        ids: &[NativeId],
        comments: &PageRequest,
    ) -> Vec<Result<Option<TaskDetailRead>, SourceError>> {
        // A page past the first is a page of one task's comments, which the detail read of it
        // alone answers.
        if comments.cursor.is_some() {
            return self.get_task_details(ids, Some(comments)).await;
        }
        if let Err(error) = validate_page(comments) {
            return ids.iter().map(|_| Err(error.clone())).collect();
        }
        let mut read = Vec::with_capacity(ids.len());
        for batch in ids.chunks(COPY_DETAIL_BATCH) {
            match self
                .graphql(
                    graphql::ISSUE_COPY_DETAILS,
                    copy_detail_batch(batch, comments, false),
                )
                .await
            {
                Ok(data) => {
                    for (slot, id) in batch.iter().enumerate() {
                        let read_one = match data.get(format!("i{slot}")) {
                            Some(node) => self.detail_of(id, node, true, None).await,
                            None => Err(SourceError::Malformed {
                                message: format!(
                                    "GitHub answered a batch read with no item for {}",
                                    id.0
                                ),
                            }),
                        };
                        read.push(read_one);
                    }
                }
                Err(error) if unresolvable_node(&error) => {
                    for id in batch {
                        read.push(self.issue_detail(id, comments).await);
                    }
                }
                Err(error) => read.extend(batch.iter().map(|_| Err(error.clone()))),
            }
        }
        read
    }

    /// Every id's item, whatever kind it is, read [`COPY_DETAIL_BATCH`] at a time with
    /// [`graphql::ISSUE_COPY_DETAILS`] — the document a copy reads the tasks it carries with —
    /// and held for the next read of each by id; see `read_ahead`.
    ///
    /// What this run wrote is never read: its own record answers, as it does for a read by id.
    /// Nor is an item a listing of its project carried whole earlier in this command — see
    /// `carried` — which is held in place of reading it.
    /// A batch GitHub refuses because one of its ids resolves to no node at all holds nothing,
    /// so each of its ids is read on its own and answered as that read answers; a draft, and an
    /// answer naming another issue than the one asked for, are left to their own reads too.
    async fn read_ahead_for_copy(&self, ids: &[NativeId]) -> Result<(), SourceError> {
        // The origins asked just before go out now, beside the members of the one project this
        // names when it names one; a project those origins answered whole is held for its next
        // read by id rather than read again here.
        let hinted = match ids {
            [one] => Some(one),
            _ => None,
        };
        self.send_pending_origins(hinted).await?;
        if let Some(project) = hinted {
            let answered = self.resolved_cache()?.get(project).cloned();
            if let Some(answered) = answered
                && answered.kind == BoardKind::Work(ItemKind::Project)
            {
                self.read_ahead()?.insert(project.clone(), Some(answered));
                return Ok(());
            }
        }
        let mut unread: Vec<NativeId> = Vec::new();
        // Read whole by a listing of its project in this command, which a mutation naming it
        // would have dropped: held as a batch would have held it, once every batch is answered.
        let mut listed: Vec<NativeId> = Vec::new();
        {
            let created = self.created()?;
            let held = self.read_ahead()?;
            let carried = self.carried()?;
            for id in ids {
                if unread.contains(id)
                    || listed.contains(id)
                    || held.contains_key(id)
                    || created.iter().any(|own| own.id == *id)
                {
                    continue;
                }
                // An issue a listing carried without the whole of what blocks it is read in a
                // batch too, which carries that, rather than having its dependency read sent
                // on its own once the copy reconciles its edges.
                let whole = carried.get(id).is_some_and(|item| {
                    item.content_kind != ContentKind::Issue || item.blocked_by.is_some()
                });
                if whole {
                    listed.push(id.clone());
                } else {
                    unread.push(id.clone());
                }
            }
        }
        // Nothing of an item's comments is wanted here: the smallest page keeps the read to
        // what the item itself costs.
        let comments = PageRequest {
            cursor: None,
            limit: 1,
        };
        // Held only once every batch has been answered, so a read-ahead that fails part way
        // holds nothing — what a listing carried included, which stays carried — and every
        // item is read on its own, as the method promises.
        let mut read: Vec<(NativeId, Option<Resolved>)> = Vec::new();
        // The issues the first id names as sub-issues, as `ISSUE_COPY_DETAILS`'s `children`
        // answered them: what a project copy's orphan check asks, when that first id is its
        // project.
        let mut children: Option<(NativeId, Vec<(String, String)>)> = None;
        for (index, batch) in unread.chunks(COPY_DETAIL_BATCH).enumerate() {
            let data = match self
                .graphql(
                    graphql::ISSUE_COPY_DETAILS,
                    copy_detail_batch(batch, &comments, index == 0),
                )
                .await
            {
                Ok(data) => data,
                Err(error) if unresolvable_node(&error) => continue,
                Err(error) => return Err(error),
            };
            for (slot, id) in batch.iter().enumerate() {
                let Some(node) = data.get(format!("i{slot}")) else {
                    continue;
                };
                let held = if node.is_null() {
                    None
                } else {
                    match optional_str(node, "__typename")? {
                        Some("DraftIssue") => continue,
                        Some("Issue") if required_str(node, "id")? != id.0 => continue,
                        _ => {}
                    }
                    // Remembered below with the rest, so a later batch refused leaves no
                    // dependency read answered from this one either.
                    self.resolve_issue_as(node, false).await?
                };
                read.push((id.clone(), held));
            }
            if index == 0
                && let Some(node) = data.get("children").filter(|node| !node.is_null())
                && node.get("id").and_then(Value::as_str) == batch.first().map(|id| id.0.as_str())
                && node.pointer("/subIssues/pageInfo/hasNextPage") == Some(&Value::Bool(false))
            {
                // A list GitHub did not answer is not an empty one: it would answer the project's
                // task listing with nothing, and the orphan check with no orphans.
                let listed = node
                    .pointer("/subIssues/nodes")
                    .and_then(Value::as_array)
                    .ok_or_else(|| SourceError::Malformed {
                        message: format!(
                            "GitHub answered the sub-issues of project {} with no list of them",
                            batch[0].0
                        ),
                    })?
                    .iter()
                    .map(|child| {
                        Ok((
                            required_str(child, "id")?.to_owned(),
                            required_str(child, "title")?.to_owned(),
                        ))
                    })
                    .collect::<Result<Vec<_>, SourceError>>()?;
                children = Some((batch[0].clone(), listed));
            }
        }
        // The project's tasks are known from this read alone when every child it named is held
        // from it, but for a document, which its title says is one and which a task listing
        // never answers: then a listing of the project's tasks is answered from them.
        if let Some((project, listed)) = children
            && read.iter().any(|(id, held)| {
                *id == project
                    && held
                        .as_ref()
                        .is_some_and(|held| held.kind == BoardKind::Work(ItemKind::Project))
            })
        {
            let mut tasks = Vec::new();
            let mut whole = true;
            for (child, title) in &listed {
                match read.iter().find(|(id, _)| id.0 == *child) {
                    Some((_, Some(held))) if held.parent.as_ref() == Some(&project) => {
                        if held.kind == BoardKind::Work(ItemKind::Task) {
                            tasks.push(held.clone());
                        }
                    }
                    _ if title.starts_with(DESIGN_TITLE_PREFIX) => {}
                    _ => whole = false,
                }
            }
            if whole {
                self.task_children()?.insert(project, tasks);
            }
        }
        let mut resolved = self.resolved_cache()?;
        let mut ahead = self.read_ahead()?;
        let mut carried = self.carried()?;
        for item in read.iter().filter_map(|(_, held)| held.as_ref()) {
            resolved.insert(item.id.clone(), item.clone());
        }
        // Taken only if still carried: a mutation naming it while the batches ran dropped it,
        // and its next read is then its own.
        for id in &listed {
            let Some(item) = carried.remove(id) else {
                continue;
            };
            // The whole of what it is blocked by came with it, so the dependency read a copy
            // makes next is answered from this one, as a copy's detail read is.
            if item.blocked_by.is_some() {
                resolved.insert(item.id.clone(), item.clone());
            }
            read.push((id.clone(), Some(item)));
        }
        for (id, held) in read {
            ahead.insert(id, held);
        }
        Ok(())
    }

    /// Add one comment to the task's issue, as the account the token belongs to.
    ///
    /// The author is refused before anything is sent — not even the task is read — because
    /// no answer GitHub could give would make posting under another name than the one asked
    /// for the right outcome.
    async fn add_comment(
        &self,
        task: &NativeId,
        comment: &NewComment,
    ) -> Result<Option<Comment>, SourceError> {
        if let Some(author) = &comment.author {
            return Err(SourceError::Refused {
                message: format!(
                    "source {} cannot post a comment as {author:?}: GitHub records the account \
                     the token signs in as the author of every comment; next: leave --author \
                     out, and the comment is posted as that account",
                    self.name
                ),
            });
        }
        let Some(issue) = self.commented_issue(task).await? else {
            return Ok(None);
        };
        let data = self
            .graphql(
                graphql::ADD_COMMENT,
                json!({"input":{"subjectId":issue.0,"body":comment.body.as_str()}}),
            )
            .await?;
        let subject = data
            .pointer("/addComment/subject")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub comment addition returned no subject".into(),
            })?;
        if required_str(subject, "id")? != issue.0 {
            return Err(SourceError::Malformed {
                message: "GitHub comment addition answered about another issue".into(),
            });
        }
        let added = data
            .pointer("/addComment/commentEdge/node")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub comment addition returned no comment".into(),
            })?;
        let added = comment_from(added)?;
        self.remember_commented(&issue)?;
        Ok(Some(added))
    }

    /// Every task commented on read together — [`DETAIL_BATCH`] to a
    /// [`graphql::ISSUE_DETAILS`] request — and each comment's hold decided on that read; then
    /// every comment it lets through added by aliased `addComment`s of one
    /// [`graphql::ADD_COMMENTS`] request per [`COMMENT_BATCH`]. GitHub answers a refusal of one
    /// alias at that alias's path, so it fails that comment alone.
    ///
    /// **Where GitHub would refuse the batch as a whole, each item is read on its own:** a
    /// batched node read naming an id that resolves to nothing is refused whole, so each id of
    /// that batch is read by itself, the missing one answered missing. A batch of one is the
    /// comment alone, as [`Self::add_comment`] adds it.
    async fn add_comments(
        &self,
        comments: &[onetaskgraph_plugin_api::BatchComment],
    ) -> Vec<Result<onetaskgraph_plugin_api::BatchAnswer<Comment>, SourceError>> {
        self.comments_batched(comments).await
    }

    async fn edit_comment(
        &self,
        task: &NativeId,
        comment: &NativeId,
        body: &CommentBody,
    ) -> Result<Option<Comment>, SourceError> {
        let Some(issue) = self.commented_issue(task).await? else {
            return Ok(None);
        };
        let Some(held) = self.comment_on(&issue, comment).await? else {
            return Ok(None);
        };
        // A comment a copy wrote keeps its origin through an edit: its marker is carried over.
        let body = CommentOrigin::kept(&held, body)?;
        let data = self
            .graphql(
                graphql::UPDATE_COMMENT,
                json!({"input":{"id":comment.0,"body":body.as_str()}}),
            )
            .await?;
        let edited = data
            .pointer("/updateIssueComment/issueComment")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub comment update returned no comment".into(),
            })?;
        let edited = comment_from(edited)?;
        if edited.id != *comment {
            return Err(SourceError::Malformed {
                message: "GitHub comment update returned the wrong comment".into(),
            });
        }
        self.remember_commented(&issue)?;
        Ok(Some(edited))
    }

    async fn delete_comment(
        &self,
        task: &NativeId,
        comment: &NativeId,
    ) -> Result<Option<NativeId>, SourceError> {
        let Some(issue) = self.commented_issue(task).await? else {
            return Ok(None);
        };
        if !self.comment_is_on(&issue, comment).await? {
            return Ok(None);
        }
        self.held_comments()?.remove(&issue);
        let data = self
            .graphql(graphql::DELETE_COMMENT, json!({"input":{"id":comment.0}}))
            .await?;
        // The payload says nothing about the comment it removed, so what is checked is that
        // GitHub answered the mutation at all rather than leaving it unanswered.
        data.get("deleteIssueComment")
            .filter(|value| !value.is_null())
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub comment deletion returned no payload".into(),
            })?;
        Ok(Some(comment.clone()))
    }

    /// Every request this source has recorded, and what each of GitHub's two budgets was
    /// attributed — read off the same accounting the session report is rendered from, so
    /// the two cannot count one request two ways.
    async fn metering(&self) -> Result<Option<Metering>, SourceError> {
        Ok(Some(self.ledger.snapshot().metering()))
    }

    /// Drop every item, search answer and board read this source holds, so the next command
    /// reads the board as a person has since left it.
    ///
    /// Every one of those is held on the assumption that nothing but this source writes the
    /// board while a command runs, which stops being true the moment the command is over: a
    /// body a person edited would be overwritten from the record held here, and a card they
    /// moved would be read as still where this source left it. The board's own field
    /// definitions go too, because a person can add or delete a `Status` option and a write
    /// resolved against the held list would not re-read on a miss. What stays is what stays
    /// valid in normal use: each repository's node id, which a miss re-reads, the pacing of
    /// mutations, which is about GitHub's limiter rather than anybody's work, and the running
    /// accounting [`metering`](TaskSource::metering) answers from.
    ///
    /// Infallible in practice: a lock an earlier failure poisoned is cleared rather than
    /// refused, because clearing it is what puts it right.
    async fn end_command(&self) -> Result<(), SourceError> {
        fn clear<T: Default>(held: &Mutex<T>) {
            *held
                .lock()
                .unwrap_or_else(std::sync::PoisonError::into_inner) = T::default();
            held.clear_poison();
        }
        clear(&self.created);
        clear(&self.updated);
        clear(&self.commented);
        clear(&self.board_cache);
        clear(&self.search_cache);
        clear(&self.narrowed_cache);
        clear(&self.search_next);
        clear(&self.resolved_cache);
        clear(&self.read_ahead);
        clear(&self.pending_origins);
        clear(&self.member_classes);
        clear(&self.carried);
        clear(&self.held_cache);
        clear(&self.children_cache);
        clear(&self.task_children);
        clear(&self.written);
        clear(&self.fields_cache);
        clear(&self.held_comments);
        clear(&self.task_writes);
        clear(&self.written_tokens);
        self.forget_at_end_of_call();
        self.expected.end_command();
        Ok(())
    }
}

/// One member of a project as a classification-only read answers it: its classification and
/// the repositories its work changes.
type MemberClass = (Classification, Vec<Repository>);

/// Each project's sub-issues as one command read them, keyed by the selector they were asked
/// for under, beside the project that selector named; see `children_cache`.
type ProjectChildren = BTreeMap<NativeId, (NativeId, Vec<Resolved>)>;

/// One issue comment as the contract carries it.
///
/// `author` is absent both when GitHub answers `null` for an account that no longer exists
/// and when it answers an actor with no login, because either way the source did not say who
/// wrote it — which is what an absent author means, rather than an author called nothing.
///
/// A comment a copy wrote here carries its origin as the hidden marker at the end of its body
/// — GitHub sets every comment's author and times itself, so there is nowhere else to keep it —
/// and is read back with the marker taken out and reported as its origin.
fn comment_from(value: &Value) -> Result<Comment, SourceError> {
    let actor = value.get("author").unwrap_or(&Value::Null);
    // A bot's login is spelled as GitHub's own pages spell it, `<login>[bot]`, so a reader —
    // and every copy of the comment — can tell it from a person's.
    let bot = actor.get("__typename").and_then(Value::as_str) == Some("Bot");
    Ok(Comment {
        origin: None,
        id: NativeId(required_str(value, "id")?.to_owned()),
        author: optional_str(actor, "login")?.map(|login| {
            if bot && !login.ends_with("[bot]") {
                format!("{login}[bot]")
            } else {
                login.to_owned()
            }
        }),
        created_at: optional_time(value, "createdAt")?,
        updated_at: optional_time(value, "updatedAt")?,
        body: required_str(value, "body")?.to_owned(),
        url: optional_str(value, "url")?.map(str::to_owned),
    }
    .unmarked())
}

/// The page of comments one issue node carries, resumed from `after`.
fn comment_page(
    node: &Value,
    issue: &str,
    after: Option<&str>,
) -> Result<Page<Comment>, SourceError> {
    let connection = node
        .get("comments")
        .filter(|value| !value.is_null())
        .ok_or_else(|| SourceError::Malformed {
            message: format!("GitHub issue {issue} answered with no comments connection"),
        })?;
    let items = optional_nodes(Some(connection), "issue comments")?
        .into_iter()
        .flatten()
        .map(comment_from)
        .collect::<Result<Vec<_>, _>>()?;
    let next = next_cursor(connection)?;
    if let Some(next) = &next {
        validate_cursor_progress(after, &next.0)?;
    }
    Ok(Page { items, next })
}

/// The far ends of an issue's whole `blockedBy` connection, when the read carried it to its
/// end — `None` when it carried none, or a page with more past it.
fn carried_blocked_by(content: &Value) -> Result<Option<Vec<Value>>, SourceError> {
    let Some(connection) = content.get("blockedBy").filter(|value| !value.is_null()) else {
        return Ok(None);
    };
    if next_cursor(connection)?.is_some() {
        return Ok(None);
    }
    Ok(Some(
        optional_nodes(Some(connection), "blocked-by issues")?
            .into_iter()
            .flatten()
            .cloned()
            .collect(),
    ))
}

/// Where the recorded tail of a dependency walk resumes; see
/// [`GitHubProjectsSource::recorded_edges`].
const RECORDED_CURSOR: &str = "onetaskgraph.depends_on:";

/// The board text field this source keeps a copy's origin in.
///
/// Named after the key it holds, and held to that name by the guard below rather than by
/// a reader noticing.
const ORIGIN_FIELD: &str = "onetaskgraph.origin";

/// The metadata key that field holds.
///
/// The engine owns this key and spells it once as `GlobalId::ORIGIN_KEY`; a plugin never
/// constructs or interprets the qualified id it carries. This source names it only to
/// route it — a short, typed value belongs in a typed field rather than in the body slot
/// a caller's own prose shares.
///
/// Restated rather than imported, because no plugin crate may depend on the engine. What
/// keeps the two spellings one contract is `scripts/check-origin-key-spelling.sh`, a
/// target in `check`: it reads the engine's own literal and fails naming the file and the
/// line when a plugin's parts from it either way. Drift here has one symptom — a copy
/// that creates a second item every run instead of finding the one it wrote — and that is
/// too late to learn it.
const ORIGIN_KEY: &str = "onetaskgraph.origin";

/// Where a recorded tail resumes, refusing a cursor no walk in `direction` reported.
///
/// The reserved key holds forward edges and nothing else — the reverse of a recorded edge
/// is derived from the far end, never written down on the near item — so only a forward
/// walk ever reports one of these cursors. A reverse read carrying one is resuming a walk
/// it did not come from, and it is told so rather than answered with an empty page that
/// reads as a walk which ended.
fn recorded_offset(
    cursor: Option<&str>,
    direction: Direction,
) -> Result<Option<usize>, SourceError> {
    cursor
        .and_then(|cursor| cursor.strip_prefix(RECORDED_CURSOR))
        .map(|offset| {
            if direction != Direction::DependsOn {
                return Err(SourceError::Config {
                    message: format!(
                        "{RECORDED_CURSOR}{offset} resumes recorded forward edges, which a \
                         reverse dependency read never issues; resume it in the direction \
                         that reported it"
                    ),
                });
            }
            offset.parse().map_err(|_| SourceError::Config {
                message: format!("{RECORDED_CURSOR}{offset} is not a recorded-edge cursor"),
            })
        })
        .transpose()
}

/// What a dependency read's own answer says of a board draft: `None` when the item is not a
/// draft, or the answer did not carry its board item; `Some(None)` for a draft whose one board
/// item is not on this board, which this source does not hold; and `Some(Some(slot))`, the
/// draft's metadata, for one it does — its board item naming this board by its number and by
/// the id the same answer read of the board, as [`GitHubProjectsSource::draft_by_id`] decides.
fn draft_slot(
    data: &Value,
    id: &NativeId,
    project_number: u32,
) -> Result<Option<Option<BTreeMap<String, Value>>>, SourceError> {
    let Some(draft) = data
        .get("node")
        .filter(|node| node.get("__typename").and_then(Value::as_str) == Some("DraftIssue"))
    else {
        return Ok(None);
    };
    let Some(items) = data.pointer("/draft/projectV2Items") else {
        return Ok(None);
    };
    if required_str(draft, "id")? != id.0 {
        return Err(SourceError::Malformed {
            message: format!("GitHub answered a different draft for {}", id.0),
        });
    }
    let nodes = items
        .get("nodes")
        .and_then(Value::as_array)
        .ok_or_else(|| SourceError::Malformed {
            message: format!("GitHub draft {} projectV2Items.nodes is not an array", id.0),
        })?;
    let info = items
        .get("pageInfo")
        .ok_or_else(|| SourceError::Malformed {
            message: format!("GitHub draft {} projectV2Items has no pageInfo", id.0),
        })?;
    if required_bool(info, "hasNextPage")? || nodes.len() > 1 {
        return Err(SourceError::Malformed {
            message: format!(
                "GitHub draft {} reports more board items than the one GitHub links a draft to",
                id.0
            ),
        });
    }
    let board = data
        .pointer("/board/projectV2/id")
        .and_then(Value::as_str)
        .ok_or_else(|| SourceError::Malformed {
            message: format!(
                "GitHub answered draft {} without this board's own id beside it",
                id.0
            ),
        })?;
    let mut held = false;
    for node in nodes {
        let project = node.get("project").ok_or_else(|| SourceError::Malformed {
            message: format!("GitHub draft {} board item has no project", id.0),
        })?;
        let number = project
            .get("number")
            .and_then(Value::as_u64)
            .ok_or_else(|| SourceError::Malformed {
                message: format!(
                    "GitHub draft {} board item has no numeric project number",
                    id.0
                ),
            })?;
        let project_id = required_str(project, "id")?;
        held |= number == u64::from(project_number) && project_id == board;
    }
    if !held {
        return Ok(Some(None));
    }
    Ok(Some(Some(
        metadata_body(optional_str(draft, "body")?.map(str::to_owned))?.1,
    )))
}

fn recorded_page(edges: Vec<DependencyEdge>, offset: usize, limit: usize) -> Page<DependencyEdge> {
    let mut page = offset_page(edges, offset, limit.max(1));
    page.next = page
        .next
        .map(|cursor| Cursor(format!("{RECORDED_CURSOR}{}", cursor.0)));
    page
}

/// The kind of one issue reached through a dependency connection.
///
/// The same questions the board scan asks, over the fields the dependency document
/// selects, and in the same order: the design prefix first, then a sub-issue is a task,
/// then anything with sub-issues or the marker is a project.
///
/// # Errors
///
/// A far end this board holds as a document is refused rather than reported. The two
/// answers that are not refusals would both be wrong: reporting it as a task names an id
/// no task read of this source can find, and reporting it as a project names one no
/// project read can. There is no third value to return — `ItemKind` has no document
/// variant, because nothing may point at a document — so the relationship itself is what
/// the person is told about.
fn related_kind(value: &Value, meaning: SubIssues) -> Result<ItemKind, SourceError> {
    let id = required_str(value, "id")?;
    if required_str(value, "title")?.starts_with(DESIGN_TITLE_PREFIX) {
        return Err(SourceError::Refused {
            message: format!(
                "GitHub issue {id} is a document of this board — its title begins \
                 {DESIGN_TITLE_PREFIX:?} — and nothing may depend on a document or be depended \
                 on by one; next: remove that issue's blocking relationship on this board"
            ),
        });
    }
    let parent = optional_str(value.get("parent").unwrap_or(&Value::Null), "id")?;
    if parent.is_some() || meaning == SubIssues::Parent {
        return Ok(ItemKind::Task);
    }
    let (_, slot) = metadata_body(optional_str(value, "body")?.map(str::to_owned))?;
    let marked = ItemKind::from_metadata(&slot).map_err(|message| SourceError::Malformed {
        message: format!("GitHub issue {id}: {message}"),
    })?;
    let sub_issues = sub_issue_total(value)?;
    Ok(if sub_issues > 0 || marked == Some(ItemKind::Project) {
        ItemKind::Project
    } else {
        ItemKind::Task
    })
}

/// The `stateInput` that closes an issue for `reason`.
fn closed_input(reason: ClosedState) -> Value {
    json!({"value":"CLOSED","stateReason":reason.reason()})
}

/// The `IssueStateUpdateInput` one status target asks for.
///
/// `stateInput` and `state` are mutually exclusive on `UpdateIssueInput`, and only this
/// one is ever sent. A non-terminal status always asks for `OPEN`, which is what reopens
/// a currently-closed issue: without that the item would read back `Unknown` and a copy
/// would report a change forever. A document has no status at all, and asks for neither.
fn state_input(target: Option<&StatusTarget>) -> Value {
    match target {
        Some(StatusTarget::Terminal(_, reason)) => closed_input(*reason),
        Some(StatusTarget::Column(_) | StatusTarget::Disabled(_)) => json!({"value":"OPEN"}),
        // A document has no status, so a write of one says nothing about the issue's open
        // or closed state rather than forcing it open: `stateInput` is what carries that
        // instruction, and an explicit null asks for no change to it.
        None => Value::Null,
    }
}

/// The metadata one write stores in the item's body slot.
///
/// The typed fields travel as themselves, so the three reserved keys are rebuilt here
/// rather than carried: the kind marker so an empty project stays readable, the
/// repository list only when it is not exactly the issue's own repository, and the far
/// ends no relationship here can name.
///
/// The copy origin is the one typed field that is also mirrored here, and only as a
/// mirror: it lands in the board's origin field as well, which stays the one every reader
/// takes it from, and it is here so that GitHub's issue search — which indexes this comment
/// and catches up with a write in seconds rather than minutes — can find the item by it.
/// A reader of the release before this one drops the slot's copy and reads the field, so an
/// item written here still reads with exactly one origin there.
fn slot_metadata(
    incoming: &Incoming<'_>,
    own_repository: Option<&Repository>,
    fallback: &[DependencyEdge],
) -> BTreeMap<String, Value> {
    let mut metadata = incoming.metadata.clone();
    match metadata.remove(ORIGIN_KEY) {
        Some(Value::String(origin)) if !origin.is_empty() => {
            metadata.insert(ORIGIN_KEY.to_owned(), Value::String(origin));
        }
        _ => {}
    }
    match incoming.written.kind() {
        BoardKind::Work(kind) => metadata.insert(
            ItemKind::METADATA_KEY.to_owned(),
            Value::String(kind.marker().to_owned()),
        ),
        // A document is told by its title, so it carries no kind marker: that key names
        // what a dependency endpoint points at, and nothing may point at a document.
        BoardKind::Document => metadata.remove(ItemKind::METADATA_KEY),
    };
    incoming.classification.record(&mut metadata);
    let derivable = own_repository
        .map(|own| incoming.repositories == [own.clone()])
        .unwrap_or(incoming.repositories.is_empty());
    if derivable {
        metadata.remove(Repository::METADATA_KEY);
    } else {
        metadata.insert(
            Repository::METADATA_KEY.to_owned(),
            Value::Array(
                incoming
                    .repositories
                    .iter()
                    .map(|repository| Value::String(repository.as_str().to_owned()))
                    .collect(),
            ),
        );
    }
    // The typed lists are what land, whatever the caller's own metadata held under their
    // keys: a key of either name travelling beside the field would otherwise be a second
    // answer to the same question, and the field is the one the contract names.
    for (key, entries) in [
        (TaskRef::DELIVERS_KEY, incoming.delivers),
        (TaskRef::DELIVERED_BY_KEY, incoming.delivered_by),
    ] {
        set_task_list(&mut metadata, key, entries);
    }
    record_edges(&mut metadata, fallback);
    metadata
}

/// Hold the far ends no relationship here can name under [`DependencyEdge::RECORDED_KEY`] in
/// one slot's metadata, or no such key when there are none.
fn record_edges(metadata: &mut BTreeMap<String, Value>, fallback: &[DependencyEdge]) {
    if fallback.is_empty() {
        metadata.remove(DependencyEdge::RECORDED_KEY);
    } else {
        metadata.insert(
            DependencyEdge::RECORDED_KEY.to_owned(),
            Value::Array(
                fallback
                    .iter()
                    .map(|edge| {
                        serde_json::to_value(&edge.to).expect("an endpoint renders as JSON")
                    })
                    .collect(),
            ),
        );
    }
}

/// Every label one item carries, from its content's own connection and nowhere else.
///
/// There is no second place to read one from: no document this source sends selects the
/// board's built-in `Labels` field, because GitHub derives it from the content and a draft
/// cannot carry one at all. The module documentation records the three schema facts that
/// settle it.
fn labels(content: &Value) -> Result<Vec<Label>, SourceError> {
    optional_nodes(content.get("labels"), "content labels")?
        .into_iter()
        .flatten()
        .map(|v| {
            Ok(Label {
                id: NativeId(required_str(v, "id")?.to_owned()),
                name: required_str(v, "name")?.to_owned(),
                color: optional_str(v, "color")?.map(str::to_owned),
            })
        })
        .collect()
}

/// Each board field value one item holds, as the change token's facets name it: the field's
/// name, with the single-select option's name or the text. A value of a type no read here
/// selects, and one naming no field, says nothing a person can rename and is left out.
fn facet_fields(field_values: &[Value]) -> Vec<(String, String)> {
    field_values
        .iter()
        .filter_map(|value| {
            let field = value.pointer("/field/name")?.as_str()?;
            let held = value.get("name").or_else(|| value.get("text"))?.as_str()?;
            Some((field.to_owned(), held.to_owned()))
        })
        .collect()
}

/// The digest of the facets of the issue `content`, whose board item's values are
/// `field_values` — or `None` when the response did not carry, whole, every member they are read
/// from: the title, the parent, the sub-issue summary and the body its kind is read from, every
/// label, and every field value.
///
/// What a write's own response spells its token from, by exactly the rules a read takes them by.
fn facets_of(content: &Value, field_values: &Value) -> Option<change_token::FacetsDigest> {
    let ended = |connection: &Value| {
        connection.pointer("/pageInfo/hasNextPage") == Some(&Value::Bool(false))
    };
    let labels = content.get("labels").filter(|labels| ended(labels))?;
    if !ended(field_values) {
        return None;
    }
    let parent = content.get("parent")?;
    let (_, slot) = metadata_body(content.get("body")?.as_str().map(str::to_owned)).ok()?;
    let kind = BoardKind::of(
        content.get("title")?.as_str()?,
        !parent.is_null(),
        sub_issue_total(content).ok()?,
        ItemKind::from_metadata(&slot).ok()?,
    );
    let labels = self::labels(&json!({"labels": labels})).ok()?;
    let nodes = field_values.get("nodes")?.as_array()?;
    Some(change_token::Facets::new(kind.facet(), facet_fields(nodes), &labels).digest())
}

/// The definition of each board field one item's values are values of, in the shape a read
/// of the board's own `fields` gives one.
///
/// A value names its field through a fragment on that field's own type, so the type is
/// known from which kind of value it is: a single-select value's field is a
/// `ProjectV2SingleSelectField`, options and all, and a text value's is a `ProjectV2Field`.
/// A value whose field carried no id, or an empty one, says nothing usable and is left out.
fn field_definitions(field_values: &[Value]) -> Vec<Value> {
    field_values
        .iter()
        .filter_map(|value| {
            let field = value.get("field")?.as_object()?;
            field.get("id")?.as_str().filter(|id| !id.is_empty())?;
            let typename = if value.get("text").is_some() {
                "ProjectV2Field"
            } else if value.get("name").is_some() {
                "ProjectV2SingleSelectField"
            } else {
                return None;
            };
            let mut defined = field.clone();
            defined.insert("__typename".to_owned(), json!(typename));
            // Only a text field holds a text value, so the value says the field's type too.
            if typename == "ProjectV2Field" {
                defined.insert("dataType".to_owned(), json!("TEXT"));
            }
            Some(Value::Object(defined))
        })
        .collect()
}

fn text_field(field_values: &[Value], name: &str) -> Result<Option<String>, SourceError> {
    let Some(node) = field_values
        .iter()
        .find(|node| node.pointer("/field/name").and_then(Value::as_str) == Some(name))
    else {
        return Ok(None);
    };
    Ok(optional_str(node, "text")?.map(str::to_owned))
}

fn valid_github_owner(owner: &str) -> bool {
    !owner.is_empty()
        && owner.len() <= 39
        && !owner.starts_with('-')
        && !owner.ends_with('-')
        && !owner.contains("--")
        && owner
            .bytes()
            .all(|byte| byte.is_ascii_alphanumeric() || byte == b'-')
}

/// GitHub's repository-name grammar: 1-100 ASCII letters, digits, `-`, `_` or `.`, and
/// neither of the two names a path segment already means.
fn valid_github_repository_name(name: &str) -> bool {
    !name.is_empty()
        && name.len() <= 100
        && name != "."
        && name != ".."
        && name
            .bytes()
            .all(|byte| byte.is_ascii_alphanumeric() || matches!(byte, b'-' | b'_' | b'.'))
}

fn valid_environment_name(name: &str) -> bool {
    let mut bytes = name.bytes();
    bytes
        .next()
        .is_some_and(|byte| byte.is_ascii_alphabetic() || byte == b'_')
        && bytes.all(|byte| byte.is_ascii_alphanumeric() || byte == b'_')
}

/// How many sub-issues one issue has.
///
/// `Issue.subIssuesSummary` is `SubIssuesSummary!` and its `total` is `Int!`, so an
/// absent or non-integer one is a response this source cannot read — and reading it as
/// zero would classify a project as a task, which is exactly the mistake the marker
/// exists to keep from happening quietly.
fn sub_issue_total(issue: &Value) -> Result<u64, SourceError> {
    let summary = issue
        .get("subIssuesSummary")
        .ok_or_else(|| SourceError::Malformed {
            message: "GitHub issue is missing subIssuesSummary".into(),
        })?;
    summary
        .get("total")
        .and_then(Value::as_u64)
        .ok_or_else(|| SourceError::Malformed {
            message: "GitHub issue subIssuesSummary.total is not an unsigned integer".into(),
        })
}

/// One issue's own `number`.
///
/// An issue always has one: GitHub declares `Issue.number` as `Int!` and every selection of
/// an issue in this module asks for it. So a read of one that comes back without it, or
/// with something that is not an unsigned integer, is a response this source cannot read —
/// absence here is **not** "this issue has no number". A draft is the content that has
/// none, and a draft never reaches this: the caller decides on `__typename` first, the way
/// it does for `subIssuesSummary`, which `DraftIssue` equally declares nothing for.
fn issue_number(issue: &Value) -> Result<u64, SourceError> {
    issue
        .get("number")
        .and_then(Value::as_u64)
        .ok_or_else(|| SourceError::Malformed {
            message: "GitHub issue number is missing or is not an unsigned integer".into(),
        })
}

/// The `number` a creating mutation answered with, and `None` when it answered without one;
/// why a missing one is tolerated is at the call in `create_and_file_issue`.
fn created_issue_number(created: &Value) -> Result<Option<u64>, SourceError> {
    match created.get("number") {
        None | Some(Value::Null) => Ok(None),
        Some(value) => value
            .as_u64()
            .map(Some)
            .ok_or_else(|| SourceError::Malformed {
                message: "GitHub created issue number is not an unsigned integer".into(),
            }),
    }
}

fn required_str<'a>(value: &'a Value, field: &str) -> Result<&'a str, SourceError> {
    value
        .get(field)
        .and_then(Value::as_str)
        .ok_or_else(|| SourceError::Malformed {
            message: format!("GitHub response is missing string field {field}"),
        })
}

fn required_nonblank_str<'a>(value: &'a Value, field: &str) -> Result<&'a str, SourceError> {
    let found = required_str(value, field)?;
    if found.trim().is_empty() {
        return Err(SourceError::Malformed {
            message: format!("GitHub response has blank string field {field}"),
        });
    }
    Ok(found)
}

/// The slot's delimiters, which `docs/metadata.md` settles once for every source that
/// needs one — Linear spells them too, in its own description field.
///
/// Restated rather than shared, because a plugin crate depends on the contract crate and
/// nothing else of this workspace. `scripts/check-metadata-slot-encoding.sh`, a target in
/// `check`, is what keeps the two one encoding: drift is otherwise quiet, since each
/// source round-trips its own writes perfectly well under its own spelling.
const METADATA_OPEN: &str = "<!-- onetaskgraph.metadata\n";
const METADATA_CLOSE: &str = "\n-->";

/// What the composer puts between a non-empty visible body and the slot, and the one thing
/// the parser takes off the visible body when it takes the slot off — exactly once, so every
/// other trailing byte of the body comes back as it was written.
// llmlint: ignore[contracts_have_one_source_or_a_drift_gate] How a composer lays the slot after prose is this source's own; `docs/metadata.md` and its gate settle only the delimiters, and no other source declares a separator to reconcile against.
const METADATA_SEPARATOR: &str = "\n\n";

/// The visible body and the metadata slot at the end of it.
///
/// The encoding is the one `docs/metadata.md` settles for Linear, which is where its
/// reasons are. Only a comment at the very end is a slot; one in the middle is a person's
/// own content and is left alone. The visible body is everything before the slot less the
/// one [`METADATA_SEPARATOR`] the composer put there, byte for byte.
fn metadata_body(
    body: Option<String>,
) -> Result<(Option<String>, BTreeMap<String, Value>), SourceError> {
    let Some(body) = body else {
        return Ok((None, BTreeMap::new()));
    };
    let Some(slot) = slot_span(&body)? else {
        return Ok((Some(body), BTreeMap::new()));
    };
    let metadata =
        serde_json::from_str(&body[slot.encoded_start..slot.encoded_end]).map_err(|error| {
            SourceError::Malformed {
                message: format!(
                    "invalid canonical JSON in GitHub issue onetaskgraph metadata slot: {error}"
                ),
            }
        })?;
    let before = &body[..slot.start];
    let visible = before.strip_suffix(METADATA_SEPARATOR).unwrap_or(before);
    Ok(((!visible.is_empty()).then(|| visible.to_owned()), metadata))
}

/// Where the metadata slot sits in one body, as byte offsets into it.
struct SlotSpan {
    /// Where [`METADATA_OPEN`] begins.
    start: usize,
    /// Where the encoded JSON begins, just past [`METADATA_OPEN`].
    encoded_start: usize,
    /// Where the encoded JSON ends, at the start of [`METADATA_CLOSE`].
    encoded_end: usize,
    /// Just past [`METADATA_CLOSE`].
    end: usize,
}

/// The slot at the very end of `body`, or `None` when it has none.
///
/// The one reading of *where the slot is*, shared by [`metadata_body`], which reads it, and
/// [`with_slot`], which rewrites it — so the two cannot disagree about which comment is the
/// slot.
fn slot_span(body: &str) -> Result<Option<SlotSpan>, SourceError> {
    let Some(start) = body.rfind(METADATA_OPEN) else {
        return Ok(None);
    };
    let encoded_start = start + METADATA_OPEN.len();
    let Some(relative_end) = body[encoded_start..].find(METADATA_CLOSE) else {
        return Err(SourceError::Malformed {
            message: "unterminated onetaskgraph metadata slot in GitHub issue body".into(),
        });
    };
    let encoded_end = encoded_start + relative_end;
    let end = encoded_end + METADATA_CLOSE.len();
    if !body[end..].trim().is_empty() {
        return Ok(None);
    }
    Ok(Some(SlotSpan {
        start,
        encoded_start,
        encoded_end,
        end,
    }))
}

/// `body` with its metadata slot holding exactly `metadata`, and every byte outside the
/// slot as it was.
///
/// A slot that is there has its JSON replaced in place; one that becomes empty is removed
/// together with the one [`METADATA_SEPARATOR`] separating it from the prose before it. A
/// body with no slot gains one the way [`compose_body`] writes it — after that separator,
/// or alone in an empty body — and a body with no slot that is given no metadata is
/// returned as it is.
fn with_slot(body: &str, metadata: &BTreeMap<String, Value>) -> Result<String, SourceError> {
    let encoded = if metadata.is_empty() {
        None
    } else {
        Some(
            serde_json::to_string(metadata).map_err(|error| SourceError::Malformed {
                message: error.to_string(),
            })?,
        )
    };
    Ok(match (slot_span(body)?, encoded) {
        (Some(slot), Some(encoded)) => format!(
            "{}{encoded}{}",
            &body[..slot.encoded_start],
            &body[slot.encoded_end..]
        ),
        (Some(slot), None) => {
            let before = &body[..slot.start];
            format!(
                "{}{}",
                before.strip_suffix(METADATA_SEPARATOR).unwrap_or(before),
                &body[slot.end..]
            )
        }
        (None, None) => body.to_owned(),
        (None, Some(encoded)) if body.is_empty() => {
            format!("{METADATA_OPEN}{encoded}{METADATA_CLOSE}")
        }
        (None, Some(encoded)) => {
            format!("{body}{METADATA_SEPARATOR}{METADATA_OPEN}{encoded}{METADATA_CLOSE}")
        }
    })
}

/// `body` with everything before its metadata slot replaced by `content`, and the slot
/// itself kept byte for byte.
///
/// The inverse of how [`metadata_body`] splits a body: the slot, when there is one, follows
/// `content` after the one [`METADATA_SEPARATOR`] the composer puts there — or alone, when
/// `content` is empty — so a read of the result reports `content` as the visible body and
/// the slot's metadata exactly as it was.
fn with_content(body: &str, content: &str) -> Result<String, SourceError> {
    let Some(slot) = slot_span(body)? else {
        return Ok(content.to_owned());
    };
    let kept = &body[slot.start..];
    Ok(if content.is_empty() {
        kept.to_owned()
    } else {
        format!("{content}{METADATA_SEPARATOR}{kept}")
    })
}

/// Hold `entries` under `key` in one slot's metadata, or no such key when there are none.
fn set_task_list(metadata: &mut BTreeMap<String, Value>, key: &str, entries: &[TaskRef]) {
    if entries.is_empty() {
        metadata.remove(key);
    } else {
        metadata.insert(
            key.to_owned(),
            Value::Array(
                entries
                    .iter()
                    .map(|entry| Value::String(entry.as_str().to_owned()))
                    .collect(),
            ),
        );
    }
}

fn compose_body(
    content: Option<&str>,
    metadata: &BTreeMap<String, Value>,
) -> Result<Option<String>, SourceError> {
    let visible = content.unwrap_or_default();
    if metadata.is_empty() {
        return Ok((!visible.is_empty()).then(|| visible.to_owned()));
    }
    let encoded = serde_json::to_string(metadata).map_err(|error| SourceError::Malformed {
        message: error.to_string(),
    })?;
    Ok(Some(if visible.is_empty() {
        format!("{METADATA_OPEN}{encoded}{METADATA_CLOSE}")
    } else {
        format!("{visible}{METADATA_SEPARATOR}{METADATA_OPEN}{encoded}{METADATA_CLOSE}")
    }))
}

fn required_bool(value: &Value, field: &str) -> Result<bool, SourceError> {
    value
        .get(field)
        .and_then(Value::as_bool)
        .ok_or_else(|| SourceError::Malformed {
            message: format!("GitHub response is missing boolean field {field}"),
        })
}
fn optional_str<'a>(value: &'a Value, field: &str) -> Result<Option<&'a str>, SourceError> {
    match value.get(field) {
        None | Some(Value::Null) => Ok(None),
        Some(value) => value
            .as_str()
            .map(Some)
            .ok_or_else(|| SourceError::Malformed {
                message: format!("GitHub response field {field} is not a string or null"),
            }),
    }
}
fn optional_nodes<'a>(
    connection: Option<&'a Value>,
    name: &str,
) -> Result<Option<&'a Vec<Value>>, SourceError> {
    match connection {
        None | Some(Value::Null) => Ok(None),
        Some(value) => value
            .get("nodes")
            .and_then(Value::as_array)
            .map(Some)
            .ok_or_else(|| SourceError::Malformed {
                message: format!("GitHub {name}.nodes is not an array"),
            }),
    }
}
fn complete_connection(connection: &Value, name: &str, size: u32) -> Result<(), SourceError> {
    let page_info = connection
        .get("pageInfo")
        .ok_or_else(|| SourceError::Malformed {
            message: format!("GitHub {name} has no pageInfo"),
        })?;
    if required_bool(page_info, "hasNextPage")? {
        return Err(SourceError::Malformed {
            message: format!(
                "GitHub {name} exceeds the supported nested connection size of {size}"
            ),
        });
    }
    Ok(())
}
fn optional_time(value: &Value, field: &str) -> Result<Option<DateTime<Utc>>, SourceError> {
    optional_str(value, field)?
        .map(|timestamp| {
            timestamp.parse().map_err(|error| SourceError::Malformed {
                message: format!("GitHub response field {field} is not a timestamp: {error}"),
            })
        })
        .transpose()
}
fn validate_page(page: &PageRequest) -> Result<(), SourceError> {
    if page.limit == 0 {
        Err(SourceError::Config {
            message: "page limit must be at least 1".into(),
        })
    } else {
        Ok(())
    }
}
fn next_cursor(connection: &Value) -> Result<Option<Cursor>, SourceError> {
    let page = connection
        .get("pageInfo")
        .filter(|value| value.is_object())
        .ok_or_else(|| SourceError::Malformed {
            message: "GitHub connection is missing pageInfo".into(),
        })?;
    if required_bool(page, "hasNextPage")? {
        let cursor = required_str(page, "endCursor")?;
        validate_cursor_progress(None, cursor)?;
        Ok(Some(Cursor(cursor.into())))
    } else {
        Ok(None)
    }
}
fn validate_cursor_progress(previous: Option<&str>, next: &str) -> Result<(), SourceError> {
    if next.is_empty() || previous == Some(next) {
        Err(SourceError::Malformed {
            message: "GitHub pagination cursor is empty or did not advance".into(),
        })
    } else {
        Ok(())
    }
}
/// The version of this plugin's opaque narrowing-search cursor.
pub const SEARCH_CURSOR_VERSION: u32 = 4;

#[derive(Serialize, Deserialize)]
#[serde(tag = "state", rename_all = "snake_case", deny_unknown_fields)]
enum SearchConnection {
    Initial {},
    Continuing { after: Cursor },
    Exhausted {},
}
impl SearchConnection {
    fn after(&self) -> Option<&str> {
        match self {
            Self::Continuing { after } => Some(&after.0),
            _ => None,
        }
    }
    fn exhausted(&self) -> bool {
        matches!(self, Self::Exhausted { .. })
    }
    /// Whether a cursor naming this position, `offset` rows into its page, is one this
    /// plugin could have handed out: a page is resumed only part of the way through it — an
    /// offset of a whole page or more would skip rows nobody was given — an initial page
    /// only once some of it was handed out, and an exhausted connection has no page to be
    /// part of the way through.
    fn valid_resume(&self, offset: usize) -> bool {
        let within = offset < SEARCH_PAGE_SIZE as usize;
        match self {
            Self::Initial { .. } => offset > 0 && within,
            Self::Continuing { after } => !after.0.is_empty() && within,
            Self::Exhausted { .. } => offset == 0,
        }
    }
}

/// Versioned source cursor. A zero offset and empty own-write ids are omitted.
#[derive(Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
struct SearchPosition {
    version: u32,
    connection: SearchConnection,
    /// How many rows of the page `connection` starts were already handed out.
    #[serde(default, skip_serializing_if = "is_zero")]
    offset: usize,
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    seen: Vec<NativeId>,
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    own: Vec<NativeId>,
}
impl Default for SearchPosition {
    fn default() -> Self {
        Self {
            version: SEARCH_CURSOR_VERSION,
            connection: SearchConnection::Initial {},
            offset: 0,
            seen: Vec::new(),
            own: Vec::new(),
        }
    }
}

fn is_zero(offset: &usize) -> bool {
    *offset == 0
}

fn numeric_cursor(cursor: Option<&Cursor>) -> Result<usize, SourceError> {
    cursor.map_or(Ok(0), |c| {
        c.0.parse().map_err(|_| SourceError::Config {
            message: "page cursor is invalid".into(),
        })
    })
}
fn offset_page<T>(mut items: Vec<T>, offset: usize, limit: usize) -> Page<T> {
    if offset > items.len() {
        return Page::last(vec![]);
    }
    let tail = items.split_off(offset);
    let mut selected = tail;
    let next = (selected.len() > limit).then(|| Cursor((offset + limit).to_string()));
    selected.truncate(limit);
    Page {
        items: selected,
        next,
    }
}

impl GitHubProjectsSource {
    /// The body [`Self::write_item`] would compose for `incoming`, sized in the characters
    /// [`BODY_LIMIT`] counts, without a request: its slot as `slot_metadata` writes it, every
    /// asset reference rewritten to its recorded URL or — for one not uploaded yet — to an
    /// address as long as GitHub's, and the heading markup this board's bodies carry.
    ///
    /// What the write reads rather than knows is taken as it most often lands: the issue in
    /// the one repository the item names, else the configured one, and every edge on the
    /// board's own relationships rather than in the slot.
    fn body_size(&self, incoming: &Incoming<'_>) -> onetaskgraph_plugin_api::BodySize {
        let own = match incoming.repositories {
            [named] => Some(named.clone()),
            _ => self
                .repository
                .as_ref()
                .and_then(|target| Repository::try_from(target.origin()).ok()),
        };
        let mut slot = slot_metadata(incoming, own.as_ref(), &[]);
        let content = match incoming.assets {
            Some(assets) => {
                let mut uploads = onetaskgraph_plugin_api::AssetUploads::default();
                for asset in &assets.assets {
                    let url = assets
                        .recorded_assets
                        .as_ref()
                        .and_then(|recorded| recorded.reusable(&asset.name, &asset.sha256))
                        .unwrap_or(PREDICTED_ATTACHMENT_URL)
                        .to_owned();
                    uploads.0.insert(
                        asset.name.clone(),
                        onetaskgraph_plugin_api::AssetUpload {
                            sha256: asset.sha256.clone(),
                            url,
                        },
                    );
                }
                incoming.content.map(|content| {
                    onetaskgraph_plugin_api::serve_asset_references(content, &mut slot, &uploads)
                })
            }
            None => incoming.content.map(str::to_owned),
        };
        let marked = content.as_deref().map(|content| self.marked(content));
        let size = compose_body(marked.as_deref(), &slot)
            .ok()
            .flatten()
            .map_or(0, |body| body.chars().count());
        onetaskgraph_plugin_api::BodySize {
            size: u64::try_from(size).unwrap_or(u64::MAX),
            limit: Some(BODY_LIMIT),
        }
    }

    async fn write_task_assets(
        &self,
        write: &ItemWrite<Task>,
        assets: Option<&onetaskgraph_plugin_api::AssetWrite>,
    ) -> Result<onetaskgraph_plugin_api::AssetsWritten, SourceError> {
        self.write_item(
            &self.task_incoming(write, assets)?,
            write.target.as_ref(),
            &write.depends_on,
        )
        .await
    }

    /// One task write as [`Self::write_item`] takes it, once what it delivers and its priority
    /// are known to be ones this source can write.
    fn task_incoming<'a>(
        &self,
        write: &'a ItemWrite<Task>,
        assets: Option<&'a onetaskgraph_plugin_api::AssetWrite>,
    ) -> Result<Incoming<'a>, SourceError> {
        let near = write.target.as_ref().unwrap_or(&write.item.id);
        for (key, entries) in [
            (TaskRef::DELIVERS_KEY, &write.item.delivers),
            (TaskRef::DELIVERED_BY_KEY, &write.item.delivered_by),
        ] {
            TaskRef::listed(key, near, Some(&self.name), entries.clone())
                .map_err(|message| SourceError::Refused { message })?;
        }
        if self.priorities.is_none() && write.item.priority != Priority::None {
            return Err(self.holds_no_priority());
        }
        if write.item.assignees.is_some() && !self.assignees {
            return Err(onetaskgraph_plugin_api::unwritable_field(KIND, "assignees"));
        }
        Ok(Incoming {
            written: Written::Work(ItemKind::Task, &write.item.status),
            title: &write.item.title,
            content: write.item.content.as_deref(),
            assets,
            labels: &write.item.labels,
            metadata: &write.item.metadata,
            repositories: &write.item.repositories,
            classification: write.item.classification,
            parent: write.item.project.as_ref(),
            delivers: &write.item.delivers,
            delivered_by: &write.item.delivered_by,
            priority: self.priorities.as_ref().map(|_| write.item.priority),
            assignees: write.item.assignees.as_deref(),
        })
    }
    async fn write_document_assets(
        &self,
        write: &ItemWrite<Document>,
        assets: Option<&onetaskgraph_plugin_api::AssetWrite>,
    ) -> Result<onetaskgraph_plugin_api::AssetsWritten, SourceError> {
        // A document takes part in no dependency graph, so there is no far end to write
        // natively and none to record: a caller naming one is told so rather than having it
        // stored under the reserved key, where a later read would report an edge the
        // contract says cannot exist.
        if !write.depends_on.is_empty() {
            return Err(SourceError::Refused {
                message: format!(
                    "this write names {} dependencies for a document, and a document takes \
                     part in no dependency graph; next: put the dependency on the task or \
                     project the document is about",
                    write.depends_on.len()
                ),
            });
        }
        self.write_item(
            &Incoming {
                written: Written::Document,
                title: &write.item.title,
                content: write.item.content.as_deref(),
                assets,
                labels: &write.item.labels,
                metadata: &write.item.metadata,
                repositories: &write.item.repositories,
                classification: write.item.classification,
                parent: write.item.project.as_ref(),
                delivers: &[],
                delivered_by: &[],
                priority: None,
                assignees: None,
            },
            write.target.as_ref(),
            &[],
        )
        .await
    }
}

/// What [`TaskSource::end_command`] leaves of this source's held state, asserted on the state
/// itself, for the two things no journey can observe.
///
/// The journeys in `crates/onetaskgraph-e2e/tests/e2e/end_command.rs` prove through the engine,
/// with and without the call, that a settlement, a board listing and a metadata search each
/// read afresh after it — the resolved records, the written-item overlay, the board and its
/// search, and the narrowed searches. What they cannot reach is the held field definitions,
/// because a status write naming an option a person deleted is refused the same whether or
/// not the list is held, and a poisoned lock, because nothing outside the source can panic
/// while one of its locks is held. So these assert those directly, and every other holder
/// beside them so a holder added later without a clear in the call fails here.
#[cfg(test)]
mod end_command_tests {
    use super::*;

    struct Token;

    impl SecretResolver for Token {
        fn get(&self, var: &str) -> Option<SecretString> {
            (var == "GH_PROJECTS_TOKEN").then(|| "test-token".into())
        }
    }

    fn source() -> GitHubProjectsSource {
        let config = serde_json::from_value(json!({
            "owner": "octo-org", "project_number": 7, "repository": "acme/work",
            // Nothing here is sent: the source is only built and its state inspected.
            "endpoint": "http://127.0.0.1:9/graphql",
        }))
        .expect("a usable configuration");
        GitHubProjectsSource::new(&SourceName::new("work").unwrap(), config, &Token)
            .expect("the source builds")
    }

    /// One issue as a board read answers it.
    fn resolved(source: &GitHubProjectsSource) -> Resolved {
        source
            .resolve(&json!({
                "id": "ITEM-1",
                "content": {"__typename": "Issue", "id": "I_1", "title": "Held",
                            "body": "what a person may since have edited", "state": "OPEN",
                            "stateReason": null, "url": null, "number": 1,
                            "subIssuesSummary": {"total": 0},
                            "labels": {"nodes": [], "pageInfo": {"hasNextPage": false}}},
                "fieldValues": {"nodes": [], "pageInfo": {"hasNextPage": false}},
            }))
            .expect("the item reads")
            .expect("an issue")
    }

    /// Hold something in every holder the call clears, and the repository id it keeps.
    fn fill(source: &GitHubProjectsSource) {
        let item = resolved(source);
        source.created.lock().unwrap().push(item.clone());
        source.updated.lock().unwrap().push(item.clone());
        *source.board_cache.lock().unwrap() = Some(Board {
            id: "PVT-board".into(),
            fields: json!({"nodes": []}),
            items: vec![item.clone()],
        });
        *source.search_cache.lock().unwrap() = Some(vec![item.clone()]);
        source
            .narrowed_cache
            .lock()
            .unwrap()
            .insert("status:todo".into(), vec![item.clone()]);
        source.children_cache.lock().unwrap().insert(
            NativeId("P-1".into()),
            (NativeId("P-1".into()), vec![item.clone()]),
        );
        source
            .search_next
            .lock()
            .unwrap()
            .insert("status:todo".into(), Some("cursor".into()));
        source
            .resolved_cache
            .lock()
            .unwrap()
            .insert(item.id.clone(), item);
        *source.fields_cache.lock().unwrap() = Some(BoardFields {
            id: BoardId::parse("PVT-board").unwrap(),
            fields: json!({"nodes": []}),
        });
        source
            .repository_cache
            .lock()
            .unwrap()
            .insert(RepositoryTarget::parse("acme/work").unwrap(), "R_1".into());
    }

    fn assert_dropped(source: &GitHubProjectsSource) {
        assert!(source.created().unwrap().is_empty(), "created");
        assert!(source.updated().unwrap().is_empty(), "updated");
        assert!(source.board_cache().unwrap().is_none(), "board");
        assert!(source.search_cache.lock().unwrap().is_none(), "search");
        assert!(source.narrowed_cache.lock().unwrap().is_empty(), "narrowed");
        assert!(
            source.children_cache.lock().unwrap().is_empty(),
            "project children"
        );
        assert!(
            source.search_next.lock().unwrap().is_empty(),
            "search paging"
        );
        assert!(
            source.resolved_cache().unwrap().is_empty(),
            "resolved records"
        );
        assert!(source.fields_cache().unwrap().is_none(), "board fields");
        assert_eq!(
            source.repository_cache().unwrap().len(),
            1,
            "a repository's node id stays valid and is kept"
        );
    }

    fn end(source: &GitHubProjectsSource) {
        tokio::runtime::Builder::new_current_thread()
            .build()
            .unwrap()
            .block_on(source.end_command())
            .expect("the command ends");
    }

    #[test]
    fn the_call_drops_every_item_search_and_board_read_and_keeps_repository_ids() {
        let source = source();
        fill(&source);
        end(&source);
        assert_dropped(&source);
    }

    #[test]
    fn the_call_clears_a_lock_an_earlier_failure_poisoned() {
        fn poison<T: Send>(held: &Mutex<T>) {
            std::thread::scope(|scope| {
                let _ = scope
                    .spawn(|| {
                        let _guard = held.lock().unwrap();
                        panic!("a failure while the lock is held");
                    })
                    .join();
            });
            assert!(held.is_poisoned());
        }
        let source = source();
        fill(&source);
        poison(&source.created);
        poison(&source.updated);
        poison(&source.board_cache);
        poison(&source.search_cache);
        poison(&source.narrowed_cache);
        poison(&source.search_next);
        poison(&source.resolved_cache);
        poison(&source.fields_cache);
        assert!(
            source.resolved_cache().is_err(),
            "a poisoned lock is refused before the call"
        );
        end(&source);
        assert_dropped(&source);
    }
}

#[cfg(test)]
mod unprojectable_fields_tests {
    use super::*;

    /// The field names `docs/metadata.md` says a `metadata_fields` entry may not name.
    fn documented() -> std::collections::BTreeSet<String> {
        let document = std::fs::read_to_string(
            std::path::Path::new(env!("CARGO_MANIFEST_DIR")).join("../../docs/metadata.md"),
        )
        .expect("docs/metadata.md is readable");
        let document = document.split_whitespace().collect::<Vec<_>>().join(" ");
        let lead = "a `field` equal, ignoring case, to ";
        let start = document.find(lead).expect("the refused-field sentence") + lead.len();
        let sentence = &document[start..];
        let sentence = &sentence[..sentence
            .find("; and two entries naming one field")
            .expect("the refused-field sentence ends at the duplicate rule")];
        let (own, github) = sentence
            .split_once(" or a field GitHub owns on every board (")
            .expect("the sentence names GitHub's own fields");
        own.split('`')
            .skip(1)
            .step_by(2)
            .chain(
                github
                    .strip_suffix(')')
                    .expect("GitHub's own fields close the sentence")
                    .split(", "),
            )
            .map(str::to_owned)
            .collect()
    }

    #[test]
    fn the_documented_refused_fields_are_exactly_the_ones_the_configuration_refuses() {
        let refused: std::collections::BTreeSet<String> = UNPROJECTABLE_FIELDS
            .iter()
            .map(|&name| name.to_owned())
            .collect();
        assert_eq!(documented(), refused, "the fields docs/metadata.md lists");
        let instance = SourceName::new("work").unwrap();
        for name in UNPROJECTABLE_FIELDS {
            for spelled in [name.to_owned(), name.to_uppercase()] {
                let entry = MetadataFieldConfig {
                    field: spelled.clone(),
                    key: "team.machine".to_owned(),
                    path: Vec::new(),
                };
                let refused = MetadataField::resolve(vec![entry], &instance)
                    .expect_err("an unprojectable field is refused");
                assert!(
                    refused.to_string().contains(&format!("{name:?}")),
                    "{spelled}: {refused}"
                );
            }
        }
    }
}