Crate incise_core 

incise’s core: content-addressed, byte-preserving markdown edits.

The premise, measured rather than assumed (bench/FINDINGS.md): a local small model editing markdown as text corrupts it. Tables were 19% correct with 28% data loss when edited directly; the same model driving these ops scores 60/60 with zero data loss. This crate is the half that has to be dependable, so its rules are the ones the benchmark forced:

• Addressing is semantic (§5.1) — heading paths, column values, item text. Never a line number, which is stale the moment anything above it moves.
• Edits splice bytes (§5.2). Everything outside the targeted range is byte-identical afterwards; the one exception is re-padding an aligned table, which is scoped to that table’s own lines.
• Refusals are the product (§5.3). An error::OpError message is written for a model to recover from in one turn, and 75% of them do.

bench/incise_ops.py is the differential-testing oracle (§9 criterion 7): where a comment here cites a corpus fixture, that fixture is the test.

Re-exports

pub use describe::describe_change;

pub use error::OpError;

pub use error::Result;

pub use front::find_frontmatter;

pub use front::format_path;

pub use front::parse_path;

pub use front::Entry;

pub use front::Fmt;

pub use front::FrontMatter;

pub use front::Kind;

pub use front::Seg;

pub use heading::find_sections;

pub use heading::heading_gap;

pub use heading::inert_headings;

pub use heading::Section;

pub use ops::dispatch::apply_op;

pub use ops::dispatch::OPS;

pub use ops::frontmatter::describe_frontmatter_change;

pub use ops::frontmatter::frontmatter_delete;

pub use ops::frontmatter::frontmatter_get;

pub use ops::frontmatter::frontmatter_set;

pub use ops::frontmatter::render_frontmatter;

pub use ops::frontmatter::render_frontmatter_get;

pub use ops::frontmatter::FrontKey;

pub use ops::frontmatter::FrontState;

pub use ops::section::render_section_outline;

pub use ops::section::resolve_section;

pub use ops::section::section_append;

pub use ops::section::section_delete;

pub use ops::section::section_insert;

pub use ops::section::section_outline;

pub use ops::section::section_rename;

pub use ops::section::section_replace_body;

pub use ops::section::section_set_level;

pub use ops::section::SectionAddress;

pub use ops::section::SectionEntry;

pub use ops::section::POSITIONS;

pub use ops::table::caption;

pub use ops::table::heading_path;

pub use ops::table::list_tables;

pub use ops::table::render_table_get;

pub use ops::table::render_table_list;

pub use ops::table::render_table_rows;

pub use ops::table::resolve_row;

pub use ops::table::resolve_table;

pub use ops::table::table_add_row;

pub use ops::table::table_delete_row;

pub use ops::table::table_get;

pub use ops::table::table_realign;

pub use ops::table::table_update_cell;

pub use ops::table::Position;

pub use ops::table::TableAddress;

pub use ops::table::TableEntry;

pub use ops::table::TableRows;

pub use ops::table::Values;

pub use table::find_tables;

pub use table::Table;

Modules

args

Runtime validation of tool arguments, ported from bench/incise_ops.py.

describe

What an op did to a document, derived from the document.

error

A failure whose message is handed straight back to a model.

front

Frontmatter parsing — the fourth sibling of crate::table, crate::list and crate::heading, and the port of bench/mdfront.py.

heading

Heading and section structure.

json

A JSON value, a parser for it, and Python’s repr of one.

list

Markdown list location and shape — the list family’s crate::table.

ops

One module per op family, in the build order §11 ranked by measured harm.

scan

Line-level scanning shared by every op family.

similar

Python difflib, ported: close-match ranking and the opcode-level diff.

table

GFM table location and geometry.