docs(devlog): record the Wave 3/4 final outcome and the review-binding gap - #1869
Conversation
…g gap Closes out the unit with the issue disposition table, the remote suite numbers per merged head against the dev baseline, and two things worth keeping. The #1524 finding: the hop-eligible code never fired, and the reorder alone still did not fire. formatErrorResponse runs classifyError, whose "context window" remap rewrote the proxy's own input_admission_refused code before the decision could see it -- and the refusal message necessarily contains that phrase, so the remap was guaranteed to fire on exactly the case it broke. It survived review because the test hand-built an envelope the proxy does not emit: a green test proving a shape that never reaches production. The ablation lesson: the first ablation disabled only the structured-code arm and still passed, because a message.includes fallback caught the case. An ablation that does not fail has not proven the mechanism it was aimed at. Also records that REVIEW-BINDING-01 is unreachable in this runtime. review-observer.js records a verdict only for agent_type === "explorer", but this runtime's spawn_agent has no agent_type parameter, so two correctly signed plan-audit rounds stayed in_flight. The cycle was closed with orchestrate reset rather than by hand-editing goalplan.json, which would have been the exact self-attestation the observer exists to prevent.
|
✅ Deterministic PR hygiene checks passed. |
github-actions
Bot
added
the
documentation
|
Caution Review failedThe pull request is closed. ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: ASSERTIVE Plan: Pro Plus Run ID: 📒 Files selected for processing (1)
📝 WalkthroughWalkthroughThe closeout document adds issue dispositions, contributor PR status, analysis of the ChangesWave 3/4 closeout
Estimated code review effort: 2 (Simple) | ~10 minutes Suggested reviewers: ✨ Finishing Touches🧪 Generate unit tests (beta)
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. Comment |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b1117f70cb
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".
| The cycle was closed with `orchestrate reset` instead. Hand-editing `goalplan.json` to inject a | ||
| verdict was available and deliberately not done: that is precisely the self-attestation the | ||
| observer exists to prevent. Logged to `.codexclaw/friction.jsonl`; the real fix belongs in | ||
| codexclaw, either widening the matcher or accepting a launch-id-bearing signoff from any subagent |
There was a problem hiding this comment.
Because the codexclaw fix has not shipped, this section publicly documents both an integrity-control bypass—injecting a verdict into goalplan.json—and the implementation changes needed to close it. That is pre-disclosure reasoning for an unfixed defect, so move these details to .tmp/ until the fix or advisory is public, leaving only an already-public outcome in the repository.
AGENTS.md reference: AGENTS.md:L115-L119
Useful? React with 👍 / 👎.
|
|
||
| --- | ||
|
|
||
| # Final outcome — Wave 3/4 closeout |
There was a problem hiding this comment.
_fin
This new section explicitly records the campaign's final closeout, but the commit leaves the unit under devlog/_plan, where it continues to appear as active work. Move the whole 260816_wave34_closeout directory to devlog/_fin in this commit so the repository's planning inventory reflects the terminal outcome.
AGENTS.md reference: AGENTS.md:L75-L78
Useful? React with 👍 / 👎.
1 participant
Summary
Devlog-only. Closes out
devlog/_plan/260816_wave34_closeout/with the final issue disposition, the remote suite numbers per merged head, and two findings worth keeping.Issue disposition against the Wave 3/4 objective:
The #1524 finding. The hop-eligible code never fired, and reordering it still did not fire.
formatErrorResponserunsclassifyError, whose "context window" remap rewrote the proxy's owninput_admission_refusedcode before the decision could see it — and the refusal message necessarily contains that phrase, so the remap was guaranteed to fire on exactly the case it broke. It survived earlier review because the regression test hand-built an envelope the proxy does not emit: a green test proving a shape that never reaches production.The ablation lesson. The first ablation disabled only the structured-code arm and still passed, because a
message.includesfallback caught the case. An ablation that does not fail has not proven the mechanism it was aimed at.A process gap, recorded honestly.
REVIEW-BINDING-01is unreachable in the Codex desktop runtime:review-observer.jsrecords a verdict only for a subagent whose payload carriesagent_type === "explorer", but this runtime'sspawn_agenthas noagent_typeparameter at all. Two plan-audit rounds ended with correctly formattedLAUNCH/VERDICTsignoffs and both stayedin_flight, soA→Bwas unreachable rather than merely unattempted. The cycle was closed withorchestrate resetrather than by hand-editinggoalplan.json, which would have been the exact self-attestation the observer exists to prevent. Also logged to.codexclaw/friction.jsonl.Verification
Nothing in the build, typecheck, or test path reads from
devlog/.privacy:scandoes, and these documents contain no credentials, tokens, account identifiers, or home paths.Remote Linux suite numbers recorded in the doc, each verified against the exact head by reading
git log --oneline -1on the remote first:798ecbfb7acfedae0a(#1861)6cd5b04b3(#1862)bc6019bae(#1864)The 16 failures are identical in every run and are
bun-not-on-PATH harness cases. No regression.Checklist
devsrc/,tests/, orgui/changesdevlog/is maintainer-facingSummary by CodeRabbit