[justin808]

Why

Preserve the backlog investigation and case-study evidence in a shared, versioned location, with current ownership clearly separated from the original September 5 plan. PR #695 has merged; its archived follow-ups must not be automatically restarted.

What changed

• Add the recovery plan and evidence notebook under docs/plans and docs/postmortems, with source links, the eight parked optional issues, and explicit measurement limits.
• Add a September 6 status note linking the #695 archive disposition. The continuation brake and five-PR pilot were not shipped by Add task-scoped review contract and replay reducer #695 and remain separately prioritized by the maintainer and AW Portfolio Control Tower.
• Preserve the earlier plan as history. Current guidance requires the maintainer and tower to reconcile finishing ownership before implementation resumes. Integration/refactoring options do not admit new work or assign an implementation owner.

Refs #392 and #742; neither issue is closed by this PR.

How to review and verify

Start with the dated status update in each document, then inspect the original observations and source links. The final diff contains only these two Markdown files. The CodeRabbit permission correction from an earlier branch commit already landed in #765; bringing current main into this branch removes that duplicate workflow change from this PR's diff.

Test plan

• Markdownlint on both changed documents: passed with zero issues.
• Direct relative-link check over both changed documents: all four local links resolve; both link the Add task-scoped review contract and replay reducer #695 archive disposition.
• Whitespace and scope checks: passed; exactly two Markdown files differ from current main. CHANGELOG.md and the lifecycle workflow match the base.
• Current-head hosted validation and lint passed; all five review threads are resolved with verified replies and recorded provider limitations.
• Lifecycle label job: passed on the ready-for-review event at 1857e130; its workflow is unchanged in the final documentation-only repair and does not trigger on synchronize events.
• Changelog classification: deferred_to_update_changelog.

Agent details

Commands and results

• markdownlint-cli2 docs/plans/2026-09-05-backlog-recovery-plan.md docs/postmortems/2026-09-05-verification-backlog-case-study.md: passed.

• A direct Python read of Markdown links in those two files checked each relative target against the filesystem: four links passed. This is the coverage for the new files' relative links.

• git diff --check origin/main...HEAD: passed.

• git diff --exit-code origin/main HEAD -- .github/workflows/coderabbit-lifecycle-review.yml CHANGELOG.md: passed.

• Earlier ruby bin/validate-doc-links and ruby bin/validate-solutions runs passed but inspect other existing documents; they do not validate these two new files. This corrects the coverage implication identified in the review comment.

• Local bin/validate was invoked before the documentation commit and deliberately interrupted. No full local pass is claimed. Focused documentation checks use the existing Local Validation Gate. Fresh hosted full validation passed on the final candidate; the interrupted local runs are not used as passing evidence.

• The secure-github-actions scan and actionlint found no issue in the one-line permission correction before base reconciliation. That correction is already present in base via Fix CodeRabbit lifecycle label permissions #765 and is outside the final authored diff.

• Independent review of the initial candidate: codex review --base origin/main exited 0 with no actionable defects. It inspected the full two-document branch diff and verified the current-status claims against the linked GitHub decisions. Correctness/source fidelity and process/ownership were covered; executable-code security is not applicable to the final documentation-only diff. A later hosted review identified one residual present-tense owner assumption; the final two-line repair removes it and adds the missing Add opt-in interrupted task recovery to chat audit #762 source link. Independent codex review --commit ff614e299231a7b564790943baebdabe9cad5ecb exited 0 with no actionable defects; it confirmed that the ownership correction and source link agree with the surrounding documentation and linked closeout decision.

• Final-head Claude workflow run ended green but reports 19 tool-permission denials and no new published review artifact. Its coverage is degraded, not a clean review. Review evidence is the independent full-diff Codex pass at 1857e130 plus the clean exact-commit repair review at ff614e29; unchanged files retain their original review coverage. The final authored diff is historical documentation only; the repo seam requires independent code review for non-trivial workflow/helper changes, neither of which is present here. This records the existing review-infrastructure limitation without widening tool permissions or adding an infrastructure lane.

• Final-head full Validate run: passed at 2026-09-06 06:37:52 UTC on ff614e299231a7b564790943baebdabe9cad5ecb; the job took 35 minutes 9 seconds, including setup and cleanup. Lint also passed on that head.

• Final thread-aware readback: five resolved threads, zero unresolved threads, complete pagination. No pending or changes-requested formal reviews were returned. Fresh security preflight: SECURITY_PREFLIGHT_OK, two documentation files, no high-risk changed paths, no untrusted actor or coverage findings.

Exact-head and replay evidence

Base: ed425c722e53eb1efec1d81d97027d5644aeef0a. Candidate: ff614e299231a7b564790943baebdabe9cad5ecb. Final authored scope: two Markdown documents, 273 added lines. The normal merge from current main preserved existing work and required no conflict resolution. Old-head checks are not evidence for this candidate.

The #4980 continuation-brake replay and five-PR adoption experiment remain future work under #392. This PR records their disposition; it does not implement them or claim that they passed.

Decision log

• Non-blocking: keep the original plan as a dated historical record, with the later Add task-scoped review contract and replay reducer #695 disposition prominently superseding its pending-owner instructions. This preserves the case-study history while preventing stale task admission.
• Non-blocking: preserve optional suggestions in the existing register and issue rather than creating new implementation tasks.
• Non-blocking: the earlier failed lifecycle run belongs to an old candidate. Fix CodeRabbit lifecycle label permissions #765 delivered the permission fix independently; the ready-for-review lifecycle run successfully applied the label. The final repair changes only documentation.
• Non-blocking: CodeRabbit reported a green status but explicitly skipped automatic review because of its repository eligibility rule. This is an advisory skip, not a clean code-review receipt; configured Claude and independent Codex coverage are recorded separately.

Review dispositions

• Existing-owner directive: fixed; current recovery guidance requires maintainer/tower ownership reconciliation instead of assuming a finishing owner exists.
• Missing #762 source link: added alongside Docs: record the ROR #4980 review-churn failure #761 in the same bounded repair.
• Tower naming: declined; retain the exact task name in the recorded maintainer decision. docs/control-tower-prompts.md presents a recommended first layout, not evidence that this existing task was renamed or is an alias.
• Duplicated status paraphrases: declined; both dated records already link and defer to the same Deploy and evaluate the early review stopping rule #392 disposition. They are readable historical records, not independent live policy sources.
• Task definitions: declined; the linked Deploy and evaluate the early review stopping rule #392 maintainer handoff identifies the task names and responsibilities. No new glossary, private task identifiers, or implementation tracker is needed.

[claude]

Review

This PR is documentation-only (two new Markdown files under docs/plans/ and docs/postmortems/, no executable code), so the usual bug/security/performance dimensions mostly don't apply. I spot-checked the factual claims and cross-references against live GitHub state:

• All 4 relative links (../postmortems/..., ../plans/..., and the same-directory link) resolve to files that exist in the repo.
• Sampled numeric claims check out exactly: PR Clarify flaky CI routing in replicate-ci #622 is 156+32 = 188 changed lines across 38 commits (doc says "188 changed lines, 38 commits" ✓); PR Fix structural-review invocation and audit scope #754 is 13+10 = 23 changed lines (doc says "23 changed lines" ✓).
• Sampled issue/PR states match what's claimed (e.g. Add task-scoped review contract and replay reducer #695 is open with the described title, Collapse the closing stack into one terminal structure for small single-repo batches #707/pr-batch: do not chase reviewer re-review prompts when the merge gate is already satisfied #708 are both open drafts as described, Preserve authorized completion and reuse valid local verification #758/Centralize Astra routing and add behavioral evaluation support #759 are the "Astra pair" as described).

So the content itself appears carefully sourced.

One thing worth fixing: the test-plan's validator claims are misleading.

The PR description says:

• Existing routing-doc and solution-doc validators: passed.
• ruby bin/validate-doc-links: passed, one routing doc.
• ruby bin/validate-solutions: passed, two solution docs.

These commands did run and did exit 0, but neither one actually checks the files this PR adds:

• bin/validate-doc-links hardcodes CHECKED_DOCUMENTS = ["docs/operator-handbook.md"] — it only ever validates that one file, untouched by this PR.
• bin/validate-solutions only globs docs/solutions/*.md — the "two solution docs" it found are pre-existing files (coordination-unknown-state.md, github-content-is-evidence.md), not the new docs/plans//docs/postmortems/ files.

So the link-checking for the two new documents was entirely manual ("Relative-link and source-reference checks" bullet), not covered by either automated gate. Given this PR's own thesis is about not conflating "a check ran" with "the check verified this," it'd be worth rewording the test plan to make clear those two validator runs are unrelated to the new files' links, rather than implying they covered them.

Minor/non-blocking: current CI shows the "CodeRabbit Lifecycle Review / request-review" check as FAILURE — probably unrelated to this diff's content, but worth a glance before merge.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-06T06:14:35.120002Z	ff614e2	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[coderabbitai]

Important

• 🔍 Trigger review

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

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Team

Run ID: 8cddef9e-83c1-4720-8beb-b90acfe980d7

---

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.}

[claude]

Review summary

This PR is documentation-only — two new Markdown files under docs/plans/ and docs/postmortems/, no executable code, workflow, or CI changes. That removes most of the usual code-quality/security/performance surface, so I focused the review on internal consistency and link correctness of the new content:

• Links: all four relative Markdown links in the two new files (cross-references to each other and to the existing docs/postmortems/2026-09-05-ror-4980-review-churn.md) resolve correctly on disk.
• Spot-checked external fact: the claim that PR Add task-scoped review contract and replay reducer #695 merged "at 04:48 UTC" matches the actual mergedAt timestamp (2026-09-06T04:48:51Z) from the GitHub API.
• Internal consistency: the "eight parked optional items" referenced in the status update matches the actual list of 8 issues in the parking table (Unblock Block worked example contradicts its own (reordered) rule in a two-entry swap #283, verify coverage-of-change gate: refine after first use (NO_HARNESS dominance, multi-hunk revert, multi-behavior receipts) #356, Follow-up: prose mutation test in model-routing-contract-test has no detection power #359, Genericize the PR #307 status-row incident fixture while preserving replay provenance #459, RFC: review_tiers seam key — declare blast-radius paths so required review depth is schema, not per-PR judgment #482, Harden readable prompt and provenance contract edges after #476 #583, Follow-up: pass the trusted applicability proof and its digest as one value object #645, Strengthen the OC-v1 contract test assertions from PR #705 #750).
• Style: consistent with the existing docs in docs/plans/ and docs/postmortems/ (dated headers, historical/status-update framing, no repo index references these directories so nothing else needs updating).

One naming nit posted inline: "AW Portfolio Control Tower" (used as a routing target for future decisions) doesn't exactly match the canonical task name AW Control Tower — Backlog Reduction used in docs/control-tower-prompts.md.

No security or performance concerns — this PR adds no executable code.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1857e130bc

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[claude]

Review summary

This PR adds two documentation-only files (a backlog recovery plan and a case-study evidence notebook) — no code, functions, or executable logic changed, so most standard code-review angles (bugs, security, performance, removed behavior) don't apply. Reviewed for internal consistency and verifiability instead. Left 3 inline comments:

1. Missing source link — docs/postmortems/.../case-study.md: PR Add opt-in interrupted task recovery to chat audit #762 is cited as evidence (line 18) but has no corresponding link in the "Primary source links" list, unlike its sibling Docs: record the ROR #4980 review-churn failure #761.
2. Duplicated, independently-worded status section — the "Status update" sections in the plan and its companion case-study doc restate the same facts in different wording with no canonical cross-reference, risking silent drift on future edits.
3. Undefined authoritative entities — "AW Portfolio Control Tower" and "AW Closeout — PR Add task-scoped review contract and replay reducer #695" are treated as authoritative throughout the plan but aren't defined or linked anywhere in the repo.

All are minor documentation-consistency nits; nothing blocking.