feat(standards): clarify publication and preflight evidence

[MrScripty]

Owning plan and exact candidate

Plan: docs/plans/authoring-publication-evidence/plan.md, milestone A1 — Explicit publication observations and early evidence failures.
Base: main at 149ba317e9397519bc181a048ecf76aef129a69c.
Head: implementation/authoring-publication-evidence-a1 at 8260489e8e3ad6b7e170bbf7bd411bac90758abd.

Intended outcome

• Keep durable publication receipts distinct from fresh, read-only target, worktree, and index observations across apply, recovery, and later status.
• Preflight destination evidence before readiness, retain final evidence and authorization checks, and expose bounded authoring diagnostics with application-purpose redaction.
• Coordinate interface 46, result projection 8, generated consumers, and interface 45 client-upgrade dispositions while retaining request contract 6 and application/readiness persistence schemas.
• Make publication observation own descriptor exhaustion, path/root races, cancellation cleanup, and a local-only read of the canonical target ref.
• Replace the supporting-workflow test's dependency on accepted Security exposure with a test-owned unexposed standard.

Prerequisites, gates, and exclusions

The only implementation prerequisite is accepted Coding-Standards main at the base above. Local A1 acceptance passed on source candidate ecbf25cdcd8218b44e68b6adb359c7efaeea765d; current head 33a93cf5 includes the path-diagnostic repair, refreshed suite-input digests, and updated execution-ledger/PR report. The plan remains Verifying until exact-head hosted checks, the required Passeur security/lifecycle review, and accepted main integration are complete.

Acceptance matrix status: A-C1 satisfied for the tested apply/recovery/status scope, including cold readback through a replacement MCP process. A-C2 remains pending until its bound-reference matrix covers both entrypoints; A1 repairs repository-content preflight and overwritten-path rejection. A-C3 and A-C4 remain pending for A2's baseline/candidate/imported-local evidence acquisition and cold-clone verification. A-C10 remains pending for independent Passeur review and acceptance-owner disposition. A3 checkout reconciliation and the overall plan objective remain pending.

Pumas-Library acquisition/runtime plans are separate from A1 and are not prerequisites. Q1/AQ-HTTP is not ready; runtime R1/R2 remain separately gated. This PR does not qualify any Pumas gate.

Excluded: A2 evidence binding, durable evidence-bundle acquisition, reconciliation writes, Pumas consumers, and migration of existing application/readiness persistence formats.

Ownership, persistence, security, and lifecycle

Repository Git owns fresh checkout/index observation and descriptor lifecycle; Analysis owns evidence validation; the Engine owns publication projection. Existing durable receipts and live publication authorization remain authoritative. No application/readiness persistence schema changes or retained-state migration are included. Descriptor exhaustion, filesystem races, and cancellation release observation resources. Application-facing diagnostics are bounded and redact private exception text. Preflight rejects evidence and exclusions that refer to Engine receipts, the attestation registry, or suite-input manifest paths that publication will overwrite; final candidate validation remains in force.

Evidence

On source candidate ecbf25cd, the combined twelve-module A1 and interface-consumer acceptance command passed 108 tests in 427.181 seconds in the offline hash-locked Python 3.12 environment. The five affected consumer modules independently passed 24 tests in 165.995 seconds. The added diagnostics regression retains bounded printable spaces, Unicode and dot-prefixed paths, rejects control/absolute/traversal forms, and preserves application redaction. verify_repository passed 73 suites and 121 checks with both refresh_verification_inputs=true and false; the refresh updated engine.py and test_analysis.py digests, committed in ecbf25cd. tools/standards_verifier/verify.py --complete passed all 73 suites, with zero failures or blocked checks. git diff --check passed.

The suite exercises test-owned local Git repositories and controlled failure, race, cancellation, readiness, and destination-overwrite cases. Mocked boundary assertions are unit evidence, not proof of real partial-clone or hosted-service behavior. No live remote repository or hosted MCP publication was qualified.

Earlier hosted run 36648831399 failed after 612 tests on nine stale current-interface assertions and did not reach its structural-verifier step. Exact prior-head run 36651259589 also failed on the obsolete interface-45 expectations in head 95b13802. Those five test assertions have now been aligned with current interface 46. Previous exact-head run 36653728027 passed on 6921489c in 44m08s. Exact-head run 36656178893 passed on 33a93cf5 in 48m06s. Exact-head run 36660853570 was canceled after commit 8260489e superseded it. Current exact-head run 36661745679 is queued and is not counted as a pass yet.

Runtime profile and optimization

The baseline main workflow run 36574954850 took 43m14s before this A1 branch. A successful prior A1 run reported 613 Engine tests in 42m35s; the other twelve package suites together took about 70s, and the final structural verifier about 8s. The workflow runs the same broad package-test sequence followed by the complete verifier. The latest completed exact-head run before this test-only optimization took 48m06s, so the long end-to-end runtime is not explained by the added tests alone.

A cProfile run of the expanded publication lifecycle test exposed an exhaustive Markdown evidence-candidate read. The test made 495 Repository.read_file calls (7.5s cumulative under profiling); selection now reads sorted candidates only until the first acceptable file and reuses its bytes. The targeted test passed in 38.6s after the change, compared with 43.6s on A1 before it and 29.5s on the base revision (one unprofiled run per revision). The repository Git suite, including its 11 new tests, passed all 52 tests in 3.7s.

The supporting-workflow test had a separate preliminary publication of its test-owned unexposed standard. It now adds that standard to the main proposal and checks authoring/application visibility at the same accepted snapshot, removing one full propose/resolve/review/apply cycle. Its focused test passed in 43.6s after the change, versus 53.5s before consolidation (one local run each; this is not a base comparison).

Reviews and remaining acceptance

GPT-6.1 Sol high's read-only review of f676cfaf found two P2 gaps: unbounded phase-specific evidence diagnostics and evidence accepted at paths publication would overwrite. GPT-6.1 Sol medium fixed both in 01bb2d03; Sol high's follow-up found both closed and no new actionable issue. Sol high's read-only review classified all nine hosted failures as stale current-interface expectations, not intentional v45 client fixtures. Sol medium changed exactly five installed-interface assertions to 46; surrounding lifecycle, disclosure, and legacy client checks remain intact. Sol high's final check of 95b13802 found no source or scope issue. Sol high's architecture review found one P2 gap: valid spaced, Unicode and dot-prefixed repository evidence paths could lose their diagnostic identity. Sol medium fixed it in e6da2e5f; Sol high's read-only follow-up confirmed closure with no other architecture findings.

GPT-6 Luna High's bounded spec/reference review found no remaining gap in the reviewed A1 contracts. Standards Engine routing selected 30 applicable standards with no unresolved questions; the resulting obligations were checked against the source.

The independent Passeur security/lifecycle review remains outstanding. The available Passeur service is bound to Pumas-Library, so no review request was sent through that unrelated repository context.

This PR remains draft pending successful exact-head hosted checks, the required Passeur review, and maintainer-authorized integration to main. No merge is requested or performed. A2 implementation starts only after A1 is accepted and merged.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 96daa6cf-0e49-43f9-a0cb-b3ac8236d2eb

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Autopilot is currently an internal CodeRabbit preview.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}