docs(handoff): ranks 9 and 22 close by measurement; #4768's root cause retracted - #610
Merged
Merged
Conversation
… RETRACTED (Redis cache decoder
…'s path-limited commit left behind handoff_doc.py commits ONLY the handoff doc, so the refs/ half of an eviction is left uncommitted in the working tree — unsaved work one routine checkout from silent deletion. Third eviction on 2026-09-14; the other two are already in. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
…w carries the closing probe event-engine-common#13 (5da5cc6467) then civitai#4839 (d2218f54da), in that order. #4839 had to be re-pinned first: it pinned #13's BRANCH head, which a squash orphans. Verified before swapping that the trees were NOT identical and that the difference was explained rather than assumed — main had gained an Apache-2.0 LICENSE and a removed hardcoded DigitalOcean credential. 🔴 The merge is not the fix landing. The endpoint still coerced minutes after, cache-busted MISS. There is no version endpoint and Cloudflare strips build headers, so BEHAVIOUR is the only deploy signal available — rank 25 now carries that probe verbatim, including the instruction to vary `limit` every run or be reading the CDN instead of the origin. #4768 auto-closed on the merge despite a bare reference. Its body carried the refuted Meilisearch attribution, so it was amended in place with the retraction at the top and the original preserved in a <details> — a closed issue asserting a wrong cause is worse than an open one, because nobody re-reads it. Also records that Submodule Pin Guard checks pin DIRECTION, not reachability from the submodule's main, so it passed on the orphaned pin on both heads. Fourth eviction to claudedocs/refs/ to stay under the ceiling: the drift-streak RESOLVED block, whose load-bearing values (run 34848144324, 0 failing / 15 executed) are all still in rank 9's row. Doc at 64,988 B of 65,536. Edits were surgical rather than section replacements, so there is no REPLACE-drop risk to read for; the ceiling was measured by hand before committing. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Follow-up to #600, landed through
handoff_doc.pyrather than by hand.Why a second PR
#600 was written with
Edit, so the handoff write-back guard's instrument could not see it and the Stop hook blocked the turn — correctly. Both halves were true: the work was recorded, and the tool that measures it had no idea. This PR re-lands the delta throughhandoff_doc.py, which is also what enforces the byte ceiling, the durable-drop warning, theforcing:/via:gates and the ranked-list ratchet. That lesson is now a Gotcha in the doc.🔴 The tool's drop-warning cannot see a table — and it fired silently here
## State nowis a REPLACE heading. The merge diff dropped the wholeMerged this sessionSHA table — seven rows, with no durable-line warning printed. The doc already warns about this exact failure ("handoff_doc.py's durable-line warning reads prose and does not see a table, so it flagged 3 unrelated lines while 8 shas were silently in the drop set") and it happened anyway, to me, one section below the warning.Those shas are rescued into
## Gotchas— an APPEND heading, where nothing replaces them — along with thecli#591concurrent-merge lesson that rode with them.What the delta records
#601(b727a83), the ordering pinned at the sentinel and the exit code34848144324,drift=success, 0 failing steps / 15 executed#4768's root cause RETRACTED: a Redis cache decoder (isNaN(Number(v)) ? v : Number(v)), not Meilisearch. Two platform PRs open, order-dependent:event-engine-common#13beforecivitai#4839. Neither auditedcli#575R2–R4 accepted in writing;devdocs#5merged (23c6d46) after re-verifying its claim liveA
closing-condition:was added to## Goal— the doc had none, and it is frozen at round 1 so the arcs this work spawned cannot extend it.Budget
The doc hit its 65,536 B ceiling twice during the previous session and once more here. Fixed the way the playbook prescribes — evicting what has CLOSED, never an open investigation or a ruled-out theory. Three blocks moved verbatim to
claudedocs/refs/(devdocs#76, thedevdocs#80ladder, the tab-forgery block), each leaving a headline pointer. Now 65,342 B.🔴 The second commit exists because
handoff_doc.py's commit is path-limited to the handoff doc, so therefs/half of an eviction is left uncommitted — unsaved work one routinecheckoutfrom silent deletion.clawgate-task:— none written.resolveexited 5; an unknown session id also answers200with an empty array, so that zero is not a clean bill of health.