From 655785438b6b91937282d1f434baa80a6c51fc75 Mon Sep 17 00:00:00 2001 From: "claude[bot]" <41898282+claude[bot]@users.noreply.github.com> Date: Thu, 2 Jul 2026 12:56:39 +0000 Subject: [PATCH 1/2] =?UTF-8?q?memory:promote=20=E2=80=94=20sibling-metric?= =?UTF-8?q?-noise-band-audit:=20PR=20#761=20resolution=20+=20orphaned-dupl?= =?UTF-8?q?icate=20gap?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Distilled from PR #761 (closing #745), which gated precision's noise band (noiseBandPct: 20), completing the sibling-metric fix that PR #762's sibling-metric-noise-band-audit.md predicted after PR #758 fixed recall alone. New finding not yet captured: even after the root-cause fix lands, duplicate issues filed against the SAME natural key (mesh-scan:perf::) by a later scan run before the fix merged are not auto-closed — planPostMergeReconcile only reconciles issues in the merged PR's closingIssuesReferences. #742, #754, #756 share natural keys with #752/#744/#746 (all fixed by PR #761) but were never referenced by any PR and remain open. Added as step 5 + updated evidence in the existing workflow doc (memory root only; quick.json untouched, still valid). Co-Authored-By: Claude Sonnet 5 --- .../sibling-metric-noise-band-audit.md | 29 +++++++++++++++---- 1 file changed, 24 insertions(+), 5 deletions(-) diff --git a/dev-mesh/curator/memory/workflows/sibling-metric-noise-band-audit.md b/dev-mesh/curator/memory/workflows/sibling-metric-noise-band-audit.md index 07b7466..adf1e14 100644 --- a/dev-mesh/curator/memory/workflows/sibling-metric-noise-band-audit.md +++ b/dev-mesh/curator/memory/workflows/sibling-metric-noise-band-audit.md @@ -1,7 +1,7 @@ --- slug: sibling-metric-noise-band-audit status: active -provenance: "PR #758 (2026-07-02), closing #745; follows quick.json#per-metric-noise-band-llm-variance (PR #461)" +provenance: "PR #758 (2026-07-02), closing #745; PR #761 (2026-07-02), closing #746; follows quick.json#per-metric-noise-band-llm-variance (PR #461)" --- # Pattern: audit sibling metrics before landing a per-metric noise-band fix @@ -45,13 +45,32 @@ root cause — recognize both when triaging a perf-regression finding. the sibling's forward-pointer issue immediately as part of the same PR — a review comment that "this should probably get its own issue" does not stop the next mesh-scan run from auto-filing duplicate `bug` issues against the ungated sibling. +5. After the gating fix merges, sweep OPEN issues carrying the same finding marker + (``) as any issue the fix PR's `Closes #N` + references, and close the rest as duplicates too. A merged PR only auto-closes / + post-merge-reconciles the issue number(s) literally in its `Closes` line + (`planPostMergeReconcile` keys off `closingIssuesReferences`) — it does not know + about other open issues that share the same natural key. Confirmed gap: PR #761 + fixed `precision`'s band (closing #746) and its merge-commit body also names + #743/#744, but three same-key duplicates filed by a second mesh-scan run before the + fix landed — #742 (`perf:3x-disjoint:precision`, dup of #752), #754 + (`perf:6x-confusable:precision`, dup of #744), #756 (`perf:12x-confusable:precision`, + dup of #746) — were never referenced anywhere and are still OPEN after the fix + merged. They will stay open forever unless something explicitly closes them; the + fix alone only stops the *next* scan from refiling. ## Evidence -Precision perf-regression `bug` issues open concurrently after PR #758 merged without -a sibling fix: #742, #746, #752, #754, #756. +Precision perf-regression `bug` issues filed against the ungated metric, two batches +from two separate mesh-scan runs before PR #761 closed the gap: first run (10:19:40Z) +— #742, #743 (recall), #744, #745 (recall), #746; second run (11:30:14Z, same natural +keys as three first-run issues, i.e. the coalesce-by-natural-key intake guard did not +catch these as dups of an already-open issue) — #752, #754, #756. PR #758 closed #745 +(recall); PR #761 closed #746 and referenced #743/#744 in its merge body — but #742, +#754, #756 remain open post-fix (step 5). ## Provenance -PR #758 (2026-07-02), closing #745. Related: `quick.json#per-metric-noise-band-llm-variance` -(PR #461), `quick.json#mir-noise-band-reversion` (PR #477). +PR #758 (2026-07-02), closing #745; PR #761 (2026-07-02), closing #746. Related: +`quick.json#per-metric-noise-band-llm-variance` (PR #461), +`quick.json#mir-noise-band-reversion` (PR #477). From a8bc64ea500617627f9342e5b2d6c6d225584625 Mon Sep 17 00:00:00 2001 From: "github-actions[bot]" <41898282+github-actions[bot]@users.noreply.github.com> Date: Sun, 5 Jul 2026 14:53:40 +0000 Subject: [PATCH 2/2] chore: nudge GitHub mergeable recalc [janitor] PR #769 stuck in mergeStateStatus=UNKNOWN for 4416min.