T-TWZD-normalize-baseline-diff-nondeterministic-output

Status: closed/done · Impact: medium · Complexity: small

Goal

run_quality_checks.ts --diff-against-baseline exists to gate a commit only on new drift — findings present in HEAD but absent from a captured baseline. But its finding unit is the raw stdout line, and several verbs emit lines that carry ephemeral content: an absolute tmpdir fixture path (e.g. /private/tmp/xxxxxx/...) or a transient commit SHA that differs every run. Those lines never match their baseline counterparts, so the per-verb set-difference reports them as new drift even when nothing real changed. The fix is a normalization pass that masks those ephemeral tokens before the lines are line-diffed, so the gate only flags genuine, deterministic drift.

Today

Location	Role today
plugin/scripts/quality_baseline.ts	findingsFromStdout splits a verb’s stdout on \n, strips only trailing whitespace per line, drops empty lines, and returns the survivors verbatim — no token masking. diff(current, baseline) then does a plain set-difference of these verbatim lines.
plugin/scripts/run_quality_checks.ts	--diff-against-baseline <sha> mode loads <baseline-dir>/<sha>.json, recomputes current findings via findingsFromStdout, calls baselineDiff, and emits new-drift: <verb>: <line> to stderr for every line in the difference (the gating set).

An ephemeral tmpdir path or transient SHA embedded in a verb’s stdout line makes that whole line unique per run, so it is always in the set-difference and always surfaces as a false-positive new-drift finding.

Proposed

Before the per-verb set-difference is computed, every finding line — on both the current side and the baseline side — passes through a normalization pass that replaces ephemeral tokens with stable placeholders:

• absolute tmpdir paths (the OS temp root and its randomized subdirectories, e.g. /tmp/..., /private/tmp/..., $TMPDIR/...) → a fixed sentinel such as <TMPDIR>;
• bare 7-to-40-char lowercase-hex commit SHAs → <SHA>.

With both sides normalized identically, two lines that differ only in an ephemeral token collapse to the same string and cancel out of the set-difference, so they no longer surface as new drift. A line whose real content changed still differs after masking and is still reported. Because the baseline JSON on disk stores pre-normalization findings, normalization is applied symmetrically at diff time rather than rewriting the capture format.

Approach

1. Add a pure normalizeFinding(line: string): string helper to quality_baseline.ts (alongside findingsFromStdout) that applies the tmpdir-path and SHA masks in a fixed order and returns the masked line. Keep the regexes anchored to token boundaries so prose that merely contains a hex word isn’t clobbered.
2. Apply normalizeFinding to both operands of the set-difference inside diff(current, baseline) (or to the finding lists immediately before the diff), so the comparison is mask-vs-mask. Decide and document whether the emitted new-drift: line shows the masked or the original text — prefer the masked form so the gate output is itself deterministic and reviewable.
3. Confirm the symmetric-masking choice against the on-disk baseline format: baselines are captured verbatim, so normalization must happen on the baseline side at diff time too (not only on current). Verify no caller reads findings expecting the raw, unmasked line in a way this changes.
4. Add a bun:test case feeding a current finding with an ephemeral tmpdir path / SHA against a baseline finding that differs only in that token, and assert the diff is empty; add a companion case where the non-ephemeral part differs and assert the line is still reported.

Files to touch

Location	Kind	Change
plugin/scripts/quality_baseline.ts	modify	Add normalizeFinding; apply it symmetrically inside diff (or to both finding lists pre-diff) before the set-difference.
plugin/scripts/run_quality_checks.ts	modify	Ensure the --diff-against-baseline path surfaces masked finding text in the new-drift: output (only if the masking is applied at the emit boundary rather than wholly inside quality_baseline.diff).

Acceptance criteria

• AC-1: A bun:test case in which the current and baseline findings differ only in a tmpdir path or commit SHA yields an empty per-verb diff (no new-drift), where today it yields one false-positive line.
• AC-2: A bun:test case in which the non-ephemeral portion of a finding line genuinely changed still reports that line as new drift after masking (no false negatives introduced).
• AC-3: normalizeFinding is applied to both the current and baseline sides of the comparison, verified by a test that masking only one side would leave the diff non-empty.

Out of scope

• The count-summary / audit-summary-line drift dimension — verbs whose summary lines (corpus counts, OK N/N, audit totals) shift run-to-run independent of any ephemeral token — is not addressed here. That dimension is already owned by T-BCNP-quality-gate-ignores-summary-and-corpus-lines (planning/backlog); this task deliberately does not duplicate or pre-empt its approach decision. This task is scoped strictly to masking ephemeral tmpdir fixture paths and transient commit SHAs.
• Changing the on-disk baseline capture format (findings stay captured verbatim; masking is a diff-time concern).

Dependencies

• none

Discovery context

Post-mortem follow-up from T-79SM-finish-claude-harness-ts-migration (closed/done). Originally drafted on PR #214, which would have landed a schema-invalid, date-prefixed, id-less task; that draft also folded in the count-summary noise dimension now owned by T-BCNP-quality-gate-ignores-summary-and-corpus-lines. This re-mint narrows the task to the genuinely new dimension — ephemeral tmpdir paths and transient SHAs — and supersedes #214.

Dedup search (spawn-from-post-mortem)

Bullet: Gate false-positives: —diff-against-baseline flagged 10 pre-existing findings as new-drift (randomized tmpdir fixture paths, incidental PR: marker lines) — proven identical on the clean tree; the masking fix is already tracked by T-TWZD-normalize-baseline-diff-nondeterministic-output. Keywords searched: t-twzd-normalize-baseline-diff-nondeterministic-output, diff-against-baseline, false-positives, pre-existing, randomized, incidental, new-drift, identical Excluded: T-XD7M-retire-task-sort-stub Top candidates (score / status / headline):

• 46 / closed/done / T-H69K-run-quality-checks-isolates-pre-existing-drift — run_quality_checks.py only fails on drift the current branch introduced
• 14 / closed/superseded / T-SDB5-audit-entities-baseline-allow — audit_entities.py: distinguish pre-existing drift from PR-introduced drift
• 13 / closed/done / T-5VN7-pr-check-cursor-bootstrap-misses-existing-comments — pr-check cursor: pin bootstrap behavior and filter self-posted orchestrator comments
• 13 / planning/proposed / T-XC32-task-acs-phrase-against-clean-fixtures — Phrase corpus-dependent task ACs against a clean fixture tree
• 12 / planning/backlog / T-BCNP-quality-gate-ignores-summary-and-corpus-lines — quality-gate ignores baseline-shifting summary and corpus-growth lines Decision: LINKED-EXISTING Rationale: The dedup script returned SPAWNED because this task scored below threshold against the bullet’s keyword set, but the originating post-mortem bullet explicitly names T-TWZD-normalize-baseline-diff-nondeterministic-output as its tracker — and this task’s scope (masking ephemeral tmpdir fixture paths and transient SHAs in --diff-against-baseline) is an exact match for the bullet’s symptom (randomized tmpdir fixture paths and incidental marker lines flagged as new-drift). Linked to this existing tracker rather than spawning a duplicate. Linked from: T-XD7M-retire-task-sort-stub

Dedup search (spawn-from-post-mortem)

Bullet: Step 7’s baseline diff flagged 8 false-positive new-drift findings: pre-existing failing suites embed randomized tempdir paths (/tmp/claude-502/append-pr-url-*/) so the differ can’t match them run-to-run — already tracked by T-TWZD-normalize-baseline-diff-nondeterministic-output; link, don’t respawn Keywords searched: t-twzd-normalize-baseline-diff-nondeterministic-output, false-positive, append-pr-url-, pre-existing, randomized, claude-502, run-to-run, new-drift Excluded: T-VWH3-ts-index-generator-script Top candidates (score / status / headline):

• 40 / closed/done / T-H69K-run-quality-checks-isolates-pre-existing-drift — run_quality_checks.py only fails on drift the current branch introduced
• 22 / planning/draft / T-TWZD-normalize-baseline-diff-nondeterministic-output — run_quality_checks —diff-against-baseline masks ephemeral tmpdir paths and transient SHAs before line-diffing
• 14 / closed/superseded / T-SDB5-audit-entities-baseline-allow — audit_entities.py: distinguish pre-existing drift from PR-introduced drift
• 13 / closed/done / T-5VN7-pr-check-cursor-bootstrap-misses-existing-comments — pr-check cursor: pin bootstrap behavior and filter self-posted orchestrator comments
• 11 / planning/backlog / T-BCNP-quality-gate-ignores-summary-and-corpus-lines — quality-gate ignores baseline-shifting summary and corpus-growth lines Decision: SPAWNED → overridden to LINKED-EXISTING Rationale: The originating post-mortem bullet (T-VWH3-ts-index-generator-script) explicitly names this task as its tracker (“already tracked by T-TWZD…; link, don’t respawn”) and per-task inputs flagged it LINKED-EXISTING. The script scored T-TWZD at rank #2 (22) but below its spawn threshold; the bullet’s symptom (randomized tmpdir paths in failing suites defeating the run-to-run differ) is exactly this task’s scope. Linked from: T-VWH3-ts-index-generator-script

Dedup search (spawn-from-post-mortem)

Bullet: quality_baseline.ts capture, run against a working tree dirty under audited paths, polluted the baseline so audit_entities surfaced phantom new-drift on - OK enumeration lines and corpus-count headers; capturing from a clean detached origin/main worktree fixed it. Also: origin/main advanced three times during the run, forcing repeated re-baseline against a moving SHA. Keywords searched: quality_baseline, audit_entities, new-drift, baseline, working-tree, origin-main, dirty, corpus, capture, moving-SHA Decision: LINKED-EXISTING T-TWZD-normalize-baseline-diff-nondeterministic-output Rationale: The dedup script scored T-TWZD at 111 / 83 across the two baseline-related bullets from this run. Both bullets are the same false-positive class T-TWZD owns — the --diff-against-baseline line-differ flags benign, non-deterministic corpus/- OK lines as new-drift whenever the baseline SHA and the worktree corpus diverge. The corpus-count-header and - OK-enumeration noise specifically overlaps T-BCNP-quality-gate-ignores-summary-and-corpus-lines (already in T-TWZD’s related). Linked here rather than spawning a duplicate. (The check_ancestry foreign-verify-commit bullet from the same run is already owned by the in-flight T-61OI-check-ancestry-flags-stale-base — not edited here because it carries an open PR #284.) Linked from: T-7RST-ensure-ready-flags-mid-migration-corpus-assumptions

Dedup search (spawn-from-post-mortem)

Bullet: Quality gate cannot mechanically pass against any frozen baseline: bun test findings embed nondeterministic temp-dir paths (/tmp/claude-502/…) — findingsFromStdout should normalize temp paths before diffing Keywords searched: findingsfromstdout, nondeterministic, mechanically, claude-502, normalize, baseline, findings, temp-dir Excluded: T-0010 Top candidates (score / status / headline):

• 67 / closed/done / T-H69K-run-quality-checks-isolates-pre-existing-drift — run_quality_checks.py only fails on drift the current branch introduced
• 58 / planning/draft / T-TWZD-normalize-baseline-diff-nondeterministic-output — run_quality_checks —diff-against-baseline masks ephemeral tmpdir paths and transient SHAs before line-diffing
• 50 / closed/superseded / T-5X6Y-task-work-step7-explicit-baseline-dir — task-work Step 7 must pass —baseline-dir explicitly to defeat silent fallback in worktree
• 45 / planning/needs-definition / T-44OO-plugin-scripts-self-discover-project-root — Plugin scripts self-discover project root from cwd
• 37 / planning/backlog / T-BCNP-quality-gate-ignores-summary-and-corpus-lines — quality-gate ignores baseline-shifting summary and corpus-growth lines Decision: SPAWNED → overridden to LINKED-EXISTING Rationale: The bullet’s proposed fix (“findingsFromStdout should normalize temp paths before diffing”) is verbatim this task’s scope — its Goal and Proposed describe a normalizeFinding helper in quality_baseline.ts that masks ephemeral tmpdir paths (and transient SHAs) before the set-difference. The script scored T-TWZD at rank #2 (58) below its spawn threshold, but linking to the active draft tracker avoids a duplicate. The H69K top hit is closed/done and only isolates pre-existing drift; it does not normalize temp paths. Linked from: T-0010

Dedup search (spawn-from-post-mortem)

Bullet: The quality gate’s baseline diff is raw-line-based; two test files printing randomized temp-dir paths produced perpetual false-positive new-drift findings, forcing an operator decision mid-run — the extractor should normalize volatile path components so the false-positive class is closed structurally, not per-test. Keywords searched: raw-line-based, false-positive, structurally, randomized, components, perpetual, new-drift, extractor Excluded: T-0012 Top candidates (score / status / headline):

• 14 / planning/draft / T-TWZD-normalize-baseline-diff-nondeterministic-output — run_quality_checks —diff-against-baseline masks ephemeral tmpdir paths and transient SHAs before line-diffing
• 12 / closed/done / T-H69K-run-quality-checks-isolates-pre-existing-drift — run_quality_checks.py only fails on drift the current branch introduced
• 9 / planning/backlog / T-BCNP-quality-gate-ignores-summary-and-corpus-lines — quality-gate ignores baseline-shifting summary and corpus-growth lines
• 5 / closed/done / T-SIHV-task-state-frontmatter-commits-on-main-not-worktree-branch — All task-state frontmatter commits land on main; worktree branch is implementation diff only
• 4 / closed/done / T-174G-archive-records-closing-timestamp-for-post-mortem-detector — sdlc lease task archive records a closing timestamp into the archive ref Decision: LINKED-EXISTING T-TWZD-normalize-baseline-diff-nondeterministic-output Linked from: T-0012

Dedup search (spawn-from-post-mortem)

Bullet: run_quality_checks.ts —diff-against-baseline flagged spurious new-drift lines from audit count-headers and new OK audit lines that shift when entity counts change; the baseline diff should normalize count headers and treat new OK lines as non-drift Keywords searched: diff-against-baseline, run_quality_checks, count-headers, new-drift, normalize, non-drift, spurious, baseline Excluded: T-0006 Top candidates (score / status / headline):

• 94 / planning/draft / T-TWZD-normalize-baseline-diff-nondeterministic-output — run_quality_checks —diff-against-baseline masks ephemeral tmpdir paths and transient SHAs before line-diffing
• 83 / planning/needs-definition / T-44OO-plugin-scripts-self-discover-project-root — Plugin scripts self-discover project root from cwd
• 78 / closed/done / T-H69K-run-quality-checks-isolates-pre-existing-drift — run_quality_checks.py only fails on drift the current branch introduced
• 71 / closed/superseded / T-5X6Y-task-work-step7-explicit-baseline-dir — task-work Step 7 must pass —baseline-dir explicitly to defeat silent fallback in worktree
• 45 / planning/backlog / T-BCNP-quality-gate-ignores-summary-and-corpus-lines — quality-gate ignores baseline-shifting summary and corpus-growth lines Decision: LINKED-EXISTING T-TWZD-normalize-baseline-diff-nondeterministic-output Linked from: T-0006 Note: the originating bullet’s symptom (audit count-headers and new - OK lines flagged as drift when entity counts change) overlaps T-TWZD’s normalization scope; the count-summary dimension is owned by T-BCNP-quality-gate-ignores-summary-and-corpus-lines per T-TWZD’s Out-of-scope.

Dedup search (spawn-from-post-mortem)

Bullet: The Step 7 baseline gate flagged new-drift lines that are pre-existing because bun test findings are non-deterministic between runs; quality_baseline and run_quality_checks should sort and dedupe each verb findings before diffing, or treat a verb whose pass/fail status is unchanged vs baseline as non-gating regardless of line churn, so flaky failing verbs do not raise false positives. Keywords searched: run_quality_checks, non-deterministic, quality_baseline, pre-existing, non-gating, regardless, new-drift, unchanged Excluded: T-F8BP-dogfood-baseline-smoke-test Top candidates (score / status / headline):

• 57 / closed/done / T-H69K-run-quality-checks-isolates-pre-existing-drift — run_quality_checks.py only fails on drift the current branch introduced
• 47 / open/ready / T-TWZD-normalize-baseline-diff-nondeterministic-output — run_quality_checks —diff-against-baseline masks ephemeral tmpdir paths and transient SHAs before line-diffing
• 32 / planning/needs-definition / T-44OO-plugin-scripts-self-discover-project-root — Plugin scripts self-discover project root from cwd
• 25 / planning/backlog / T-BCNP-quality-gate-ignores-summary-and-corpus-lines — quality-gate ignores baseline-shifting summary and corpus-growth lines
• 22 / closed/superseded / T-5X6Y-task-work-step7-explicit-baseline-dir — task-work Step 7 must pass —baseline-dir explicitly to defeat silent fallback in worktree Decision: SPAWNED → overridden to LINKED-EXISTING Rationale: The script returned SPAWNED because the top hit (T-H69K, closed/done) only isolates pre-existing drift; but T-TWZD (rank #2, open/ready) is an exact-scope match — its findingsFromStdout normalization pass is precisely the fix for non-deterministic per-verb findings the F8BP dogfood run hit (bun test: PR: https://github.com/example/repo/pull/{1,7} flagged as new-drift while byte-identical to the baseline). This is the sixth post-mortem to surface the same gap and link here. Linked from: T-F8BP-dogfood-baseline-smoke-test Bullet: Step 7 —diff-against-baseline flagged spurious new-drift lines that were OK audit lines for other tasks, because the baseline was captured at origin/main while the worktree branched from a local main that parallel sessions had advanced. Capturing the baseline at the worktree’s actual branch-point rather than origin/main would avoid the transient false-positive. Keywords searched: diff-against-baseline, false-positive, branch-point, new-drift, capturing, transient, spurious, baseline Excluded: T-SPBO-commit-helper-routes-through-tempfile Top candidates (score / status / headline):
• 104 / open/ready / T-TWZD-normalize-baseline-diff-nondeterministic-output — run_quality_checks —diff-against-baseline masks ephemeral tmpdir paths and transient SHAs before line-diffing
• 79 / planning/needs-definition / T-44OO-plugin-scripts-self-discover-project-root — Plugin scripts self-discover project root from cwd
• 67 / closed/done / T-H69K-run-quality-checks-isolates-pre-existing-drift — run_quality_checks.py only fails on drift the current branch introduced
• 57 / closed/superseded / T-5X6Y-task-work-step7-explicit-baseline-dir — task-work Step 7 must pass —baseline-dir explicitly to defeat silent fallback in worktree
• 38 / planning/backlog / T-BCNP-quality-gate-ignores-summary-and-corpus-lines — quality-gate ignores baseline-shifting summary and corpus-growth lines Decision: LINKED-EXISTING T-TWZD-normalize-baseline-diff-nondeterministic-output Linked from: T-SPBO-commit-helper-routes-through-tempfile Note: the originating bullet’s symptom (a baseline captured at origin/main while the worktree branched from a locally-advanced main, so other tasks’ - OK audit lines surfaced as spurious new-drift) is the same false-positive class T-TWZD’s normalization pass targets; linked rather than respawning.

Dedup search (spawn-from-post-mortem)

Bullet: run_quality_checks —diff-against-baseline against audit_entities produced false-positive new-drift lines that were benign OK corpus-listing rows and a file-count header, flagged only because the worktree corpus was behind origin/main; the baseline line-diff is sensitive to corpus membership churn, not just schema drift Keywords searched: diff-against-baseline, run_quality_checks, audit_entities, false-positive, corpus-listing, file-count, membership, new-drift Excluded: T-E9C9-project-check-flags-operations-column-drift Top candidates (score / status / headline):

• 64 / open/ready / T-TWZD-normalize-baseline-diff-nondeterministic-output — run_quality_checks —diff-against-baseline masks ephemeral tmpdir paths and transient SHAs before line-diffing
• 47 / closed/done / T-H69K-run-quality-checks-isolates-pre-existing-drift — run_quality_checks.py only fails on drift the current branch introduced
• 42 / planning/needs-definition / T-44OO-plugin-scripts-self-discover-project-root — Plugin scripts self-discover project root from cwd
• 21 / closed/superseded / T-5X6Y-task-work-step7-explicit-baseline-dir — task-work Step 7 must pass —baseline-dir explicitly to defeat silent fallback in worktree
• 19 / planning/backlog / T-BCNP-quality-gate-ignores-summary-and-corpus-lines — quality-gate ignores baseline-shifting summary and corpus-growth lines Decision: LINKED-EXISTING T-TWZD-normalize-baseline-diff-nondeterministic-output Linked from: T-E9C9-project-check-flags-operations-column-drift Note: the originating bullet’s symptom (audit - OK corpus-listing rows and a file-count header surfacing as new-drift because the worktree corpus was behind origin/main) is the same baseline-diff false-positive class T-TWZD targets; linked rather than respawning. Bullet: Baseline-diff quality gate reported spurious new-drift lines for audit_entities corpus rows that differ only because the branch was cut from an older main; run_quality_checks.ts —diff-against-baseline should diff on genuine finding lines rather than the whole corpus listing Keywords searched: diff-against-baseline, run_quality_checks, audit_entities, baseline-diff, new-drift, reported, spurious, quality Excluded: T-7PXF-declare-worktree-init-bun-install Top candidates (score / status / headline):
• 81 / open/ready / T-TWZD-normalize-baseline-diff-nondeterministic-output — run_quality_checks —diff-against-baseline masks ephemeral tmpdir paths and transient SHAs before line-diffing
• 64 / closed/done / T-H69K-run-quality-checks-isolates-pre-existing-drift — run_quality_checks.py only fails on drift the current branch introduced
• 58 / planning/needs-definition / T-44OO-plugin-scripts-self-discover-project-root — Plugin scripts self-discover project root from cwd
• 47 / closed/superseded / T-5X6Y-task-work-step7-explicit-baseline-dir — task-work Step 7 must pass —baseline-dir explicitly to defeat silent fallback in worktree
• 37 / closed/done / T-T879-task-work-uses-per-project-quality-checks — Make /sdlc:task-work quality-check commands per-project configurable Decision: LINKED-EXISTING T-TWZD-normalize-baseline-diff-nondeterministic-output Linked from: T-7PXF-declare-worktree-init-bun-install Note: the symptom here (audit_entities - OK corpus rows and the count header flagged as new-drift because the branch was cut from an older main) is the count-summary/corpus-growth dimension owned by T-BCNP-quality-gate-ignores-summary-and-corpus-lines rather than T-TWZD’s ephemeral-token masking; linked here as the nearest active tracker per the dedup search, with the cross-reference to T-BCNP noted. Bullet: The baseline-gated quality gate reported new-drift lines that were all benign OK audit-status lines from audit_entities.ts, flagged only because the baseline SHA origin/main predated local-main commits; the per-entity OK lines should be excluded from the baseline diff or the baseline captured against local main tip Keywords searched: baseline-gated, audit_entities, audit-status, local-main, per-entity, new-drift, reported, baseline Excluded: T-WOL2-task-work-preflight-probes-file-mutation-tools Top candidates (score / status / headline):
• 92 / planning/needs-definition / T-44OO-plugin-scripts-self-discover-project-root — Plugin scripts self-discover project root from cwd
• 88 / open/ready / T-TWZD-normalize-baseline-diff-nondeterministic-output — run_quality_checks —diff-against-baseline masks ephemeral tmpdir paths and transient SHAs before line-diffing
• 66 / closed/done / T-H69K-run-quality-checks-isolates-pre-existing-drift — run_quality_checks.py only fails on drift the current branch introduced
• 50 / closed/superseded / T-5X6Y-task-work-step7-explicit-baseline-dir — task-work Step 7 must pass —baseline-dir explicitly to defeat silent fallback in worktree
• 39 / planning/backlog / T-BCNP-quality-gate-ignores-summary-and-corpus-lines — quality-gate ignores baseline-shifting summary and corpus-growth lines Decision: LINKED-EXISTING T-44OO → overridden to LINKED-EXISTING T-TWZD-normalize-baseline-diff-nondeterministic-output Rationale: The script’s top pick T-44OO (score 92) is about plugin scripts self-discovering the project root from cwd — a different concern. The bullet’s actual symptom (benign - OK audit lines surfacing as spurious new-drift because the baseline SHA origin/main predated the local-main commits the worktree branched from) is the exact false-positive class T-TWZD owns (rank #2, score 88), already linked from six prior post-mortems with the identical symptom. Linked to T-TWZD rather than spawning a duplicate. Linked from: T-WOL2-task-work-preflight-probes-file-mutation-tools

Post-mortem

Captured by /sdlc:task-work on 2026-06-05. PR: pending.

Acceptance criteria coverage

• AC-1: auto — bun test plugin/scripts/tests/quality_baseline.test.ts cases tmpdir-only difference cancels out and SHA-only difference cancels out: a current/baseline pair differing only in a tmpdir path or SHA yields an empty diff().
• AC-2: auto — same suite, genuine change still surfaces after masking: a real non-ephemeral change (expected 3 → expected 4) is still reported, in masked form, so no false negative is introduced.
• AC-3: auto — same suite, masking only one side would leave the diff non-empty: the symmetric path returns [], and the asymmetric counterfactual (mask current only, compare against raw baseline) leaves the masked line unmatched — proving both sides must be masked.

What worked

• The fix dogfooded cleanly: re-running the baseline-gated gate on the worktree returned OK 5/5 with 0 new-drift, and the real bun test vX (7c45ed97) banner hash in the captured baseline confirmed normalizeFinding masks an actual transient SHA, not just synthetic test strings.
• Symmetric masking landed in one place (diff() masks both operands); run_quality_checks’ pre-existing loop reused the same exported helper, so the classification stayed consistent with no duplicated regex.

Friction and automation gaps

• The verify-stamp, start-commit, and (upcoming) prs-record steps assume the main checkout sits on main; this run’s main checkout was on an unrelated user branch (backlog-capture), so ensure_ready_mutate.ts --commit-on main and start_task.ts (both resolve the main repo via --git-common-dir, then require branch==main) could not be driven directly — the run used a throwaway main-ref worktree plus start_task.ts --main-repo <override> and a hand-applied --commit for the verify-stamp. A --main-repo/--commit-on <worktree> override on the ensure-ready mutator (mirroring the one start_task.ts already exposes) would let task-work land main-ref task-state commits on a dedicated worktree without the operator improvising, when the primary checkout is intentionally parked on another branch. → T-31XR-ensure-ready-mutator-main-repo-override

Spawned follow-up tasks

• T-31XR-ensure-ready-mutator-main-repo-override (https://github.com/sksizer/dev/pull/297) — ensure-ready mutator main-repo override; spawned (Upstream-plugin/sdlc, degenerate-local).

---

← Back to Tasks