Module check 

Check records — the engine-recorded act of verification (agent-trust plan 14).

A check is an agent recording “entity E checked, verdict ok | failed, via method M”. It is engine state, never entity content: absent from markdown and content_hash, and it produces no mem commit — checking-touches-nothing is what makes check-staleness computable. Records are append-only JSONL under the workspace store (.memstead/state/checks/checks.jsonl); a newer check supersedes older ones for state derivation but never erases them.

Unlike the friction ledger next door, recording here is NOT best-effort: a check the ledger failed to persist must refuse — the caller believes the act was recorded, and a silently dropped check is exactly the self-report dishonesty this tier exists to end. For the same reason there is no rotation cap: check history is the substrate process state derives from, not disposable telemetry.

Each record carries plan-13 provenance (actor, client, declared role) plus the entity’s content_hash at check time. State derivation compares that hash against the current one:

• no record → never_checked
• hash matches, ok → checked_ok
• hash matches, failed → check_failed
• hash differs → check_stale (whatever the verdict was, it no longer speaks to the current content — stated, never silently carried forward)

Structs

CheckLedger

Append/read handle for a workspace’s check ledger.

CheckRecord

One recorded check — the full ledger line.

Enums

CheckState

Derived per-entity check state.

Verdict

A check verdict from the closed vocabulary.

Constants

VERDICTS

The closed verdict vocabulary. Nuance goes in the method note or in process-mem entities — never in new verdict values.

Functions

check_ledger_path

The ledger file path for a workspace.

derive_state

Derive the state from the newest record (if any) and the entity’s current content_hash.