amanuensis

Method

Review coverage

Which failure modes were tested where, and the disposition each one reached.

Branch  main Checked  ec5824bceb9c As of  2026-10-07 22:54:09 Stale: 0 with a survey obligation, 65 exempt, of 655 scoped files

§Coverage overview

50Recorded reviews
9Subsystems
12Active concerns
4Confirmed defects

Only the 50 recorded reviews are expanded below; 58 unassessed intersections do not consume reading space. The thin rule shows recorded coverage; the labelled markers show review outcomes. 26 reviews are linchpin-dependent.

Confirmed defectAccepted behaviorRuled outOut of scopeCompeting explanationsLinchpin-dependentNot assessed
Full cross-reference matrix9 subsystems × 12 concerns

Use the matrix when exact row-and-column comparison matters. Headers and subsystem names remain fixed while the matrix scrolls.

Full concern review matrix
SubsystemAT-1AT-2CC-1CR-1EP-1IF-1RL-1SC-1SI-1SI-2TB-1TR-1
Survey methodology and agent contractsB-01
MCP core, persistence, and lifecycleB-02
Knowledge tools and workflow APIB-03
Diff-aware materializerB-04
Packaging, installer, validation, and product docsB-05Not assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessed
Report interface design and validation studiesB-06Not assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessed
Embedded research surveys and platform trialsB-07Not assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessed
Activation evidence and release readinessB-08Not assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessed
Conspectus design lanes, their drivers and receiptsB-09Not assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessedNot assessed

§Active concerns

AT-1

Category
atomicity
Origin
seeded
Discovered in
—

Notes

T7: Inspect every multi-table mutation for one transaction covering the invariant it claims — the carry and its run counts, a disposition and its evidence join, a reconciliation and its scope_gaps rewrite.

AT-2

Category
checkpoint atomicity
Origin
seeded
Discovered in
—

Notes

T7: Mutate the database under WAL, call commit_phase_gate, clone or restore the storage commit, and verify the mutation is in it. This is the territory of archived finding B02-1 and of the A0 baseline's recoverability defect.

CC-1

Category
derived-state coherence
Origin
seeded
Discovered in
—

Notes

T2: Enumerate every derived surface — the SQL views, the materialized pages, the manifest, the search index, the dashboard census — and find one that can disagree with the rows it is derived from after an ordinary write.

CR-1

Category
concurrency
Origin
seeded
Discovered in
—

Notes

T8: Race two writers, the lock table, SQLite WAL commits and storage git commits; reconcile the landed records against the history. Lanes run concurrently against sibling worktrees on this machine.

EP-1

Category
exceptional paths
Origin
seeded
Discovered in
—

Notes

T4: Fail validation, SQL and filesystem steps inside mutating tools and compare the dependent rows before and after. The success path establishes an invariant; ask whether the error path restores it.

IF-1

Category
incremental/full parity
Origin
seeded
Discovered in
—

Notes

T6: Render incremental, force-full and clean_publish from identical state and compare state, coverage and bytes; then delete a record and re-render, and look for what the incremental path leaves behind.

RL-1

Category
resource lifecycle
Origin
seeded
Discovered in
—

Notes

T9: Inject success, error, timeout and cancellation at every database handle, child process, temporary directory and materializer lock boundary, and find one that is not released on every exit path.

SC-1

Category
seam contracts
Origin
seeded
Discovered in
—

Notes

T11: Find a fact published in two places with nothing comparing them: the enum in the vocabulary contract against the SQL CHECK and the two generated modules, the tool inventory against DEVELOPMENT.md, the server's refusal messages against the skill references that instruct the calls they refuse, a schema field against the Python projection that reads it.

SI-1

Category
scope identity
Origin
seeded
Discovered in
—

Notes

T1: Try to make project-key and workspace-binding derivation conflate two checkouts of this repository, or place a store outside the configured storage root. This repository is surveyed from git worktrees whose .amanuensis is worktree-local, so two live stores for one project_key is the ordinary case, not the exotic one.

SI-2

Category
revision identity
Origin
seeded
Discovered in
—

Notes

T1: Follow every ref_sha, first_seen, anchor and repaired_sha from the write that accepts it to the read that trusts it; require each to resolve in the bound workspace at the moment it is read, and try to reuse one at a revision the workspace cannot reach.

TB-1

Category
temporal bounds
Origin
seeded
Discovered in
—

Notes

T3: Enumerate every git and python subprocess a tool call awaits; locate the timeout, the cancellation path and the failure propagation for each. An unbounded probe on the activation path hangs the host.

TR-1

Category
trust boundary
Origin
seeded
Discovered in
—

Notes

T10: Fuzz tool arguments that reach a filesystem path, a SQL identifier or a git argument: storage escape, injection, and bypass of the depth authority the phase ladder enforces.