chore(deps): pin pr-reviewer-plugin v0.12.0 — grounding, dispositions, fan-out width - #11
Merged
Merged
Conversation
…xplained clearance v0.9.0 (pr-reviewer-plugin#27, issue #26). A zero-finding PASS is the highest-consequence verdict the panel posts: it dismisses our own standing REQUEST_CHANGES and clears the promotion path. On protoAgent#2141 the panel confirmed a major on one head, returned PASS with zero findings on the next with the code unchanged, the block lifted, and the defect merged 44 seconds later (protoAgent#2143, since fixed by #2164). A miss can't be caught the way a hallucination can — there's no claim to re-ground, and `findings=0` reads identically whether the code is clean or nobody looked. So the rule is structural: a blocker/major that DISAPPEARS without being fixed, carried, or refuted is unproven, not resolved, and the block stays up. The verdict still posts and names the dropped finding. The escape hatch matters as much as the rule: the FIRST clean PASS after a blocker/major holds, a SECOND consecutive one lifts — two independent draws finding nothing is evidence, one is a coin flip. Off switch: PR_REVIEWER_HOLD_UNEXPLAINED_CLEARANCE=false. Also gitignores data/review-eval/ — the hand-grounded findings dataset from the dogfood, kept local until it's worth moving to the protolabs node (protoLab#24). Labels and claim text only, no secrets. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…s + fan-out width Supersedes the v0.9.0 pin on this branch. v0.10.0 evidence grounding (a finding quoting absent code is downgraded to uncertain and can't carry a FAIL), v0.11.0 prior-finding dispositions (a confirmed blocker/major must be accounted for at any verdict), v0.12.0 panel fan-out width + per-step timings. NOTE the fan-out width (max_concurrency: 5 in the recipe) is a NO-OP until the base moves: it needs protoAgent#2168, which landed after the v0.107.0 tag, and this bundle pins protoagent 0.106.0. The grounding and disposition work has no such dependency and takes effect on this pin alone. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
mabry1985
changed the title
Adds v0.13.0 (a promoted WARN carries its findings + the marker gains findings=N), v0.13.1 (hermetic webhook test), v0.14.0 (dispatcher config resolved LIVE — editing repos or flipping shadow_mode/kill switches through Settings now takes effect without a restart, issue #11) on top of the v0.12.0 grounding + dispositions work. The fan-out width remains a no-op until the base moves off protoagent 0.106.0 (needs protoAgent#2168, which landed after the v0.107.0 tag). Co-Authored-By: Claude Opus 4.8 (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
1 participant
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.
Moves Vera to pr-reviewer-plugin v0.12.0 — three releases, all from one night of dogfooding the panel against protoAgent's own PRs.
What lands
v0.10.0 — evidence grounding (#25). A finding whose quoted code appears nowhere in the cited file at the reviewed head (nor in the PR's patch for it) is annotated
uncertain, whichverdict_foralready refuses to turn into a FAIL. On protoAgent#2138 the panel confirmed_writable_dir()"drops the.expanduser()call" — the head contains it verbatim — on two consecutive heads, escalating to blocker, with the operator's blob refutation and a passing test already on the PR. Fail-open throughout; nothing is ever dropped, only stripped of gating power, with the missing quote named in the body.v0.11.0 — prior-finding dispositions (#26). The report pass must state, per prior blocker/major, whether it was
fixed/open/refuted. An undispositioned one holds any standing block at any verdict. Generalizes v0.9.0's rule, which could only guard a zero-finding PASS — and protoAgent#2150 showed a major vanishing into a WARN instead. The two guards are a fallback chain: when dispositions are present they are the authority.v0.12.0 — panel fan-out width + per-step timings. The five finders are one parallel stage, but the host caps concurrency at 4, so the stage ran 4+1 and paid the slowest finder twice. Measured over 60 reviews: 5-finder p50 458s vs 4-finder 322s, solving to ~136s per finder and ~186s for the sequential tail.
max_concurrency: 5needs protoAgent#2168, which landed after thev0.107.0tag — and this bundle pins protoagent 0.106.0. So the ~136s saving needs a protoAgent release 0.108.0 → base bump → roll, as a separate change.Grounding and dispositions have no such dependency and take effect on this pin alone.
Evidence base
24 findings hand-grounded against blobs across the night: 18 true, 3 false, 1 confirmed miss, 2 unverified. The three false positives were all string/prefix predicates the panel didn't evaluate; the true ones cluster in removed-behaviour and cross-file reasoning. Tracked in protoLab#24.
Merging this deploys — GHCR rebuild then watchtower rolls Vera in ~60s.
🤖 Generated with Claude Code