🎯 Summary

#15300 did not finish the job, and this is the correction to my own claim. Its body says it "unblocks dev-refresh for every feature branch." It does not.

The moment #15300 merged (38a7285973) I went to land #15272's roster join. The sync guard passed — that half works. The merge still failed, on check-whitespace:

Trailing whitespace found in resources/content/discussions/chunk-1/discussion-15173.md:24 … :479

Pipeline-authored, carried in by the merge, never touched by the branch.

Resolves #15303 · Refs #15281, #15299

🧭 Why that whitespace is correct, and the escape isn't ours

Trailing whitespace is how markdown encodes a hard line break, so GitHub discussion bodies are full of it. The pipeline knows — its own workflow says so:

<h1 class="neo-h1" data-record-id="4">Commit (neo repo). --no-verify: generated data fails whitespace hooks.</h1>

NEO_SKIP_TICKET_ARCHAEOLOGY=1 git commit --no-verify -m "chore(data): … [skip ci]"

That escape belongs to a trusted CI job committing generated data. It does not extend to a maintainer who later inherits those commits through a merge — which is everyone. --no-verify is banned here and also skips the AiConfig test-mutation safety lint.

check-whitespace runs on lint-staged's * glob — every staged file of every type — so it was always going to inherit the same defect the moment a merge staged the pipeline's content. I scoped #15299 to the hook that bit me instead of asking whether the defect was hook-layer-wide.

🔬 The rule is extracted, not copied

#15300 put the discriminator inline. A second inline copy would be a duplicated primitive (ADR-0019 C2), and the copy that drifts is the one nobody notices — it fails open, silently, on exactly the content it exists to police. mergeInheritance.mjs now owns the rule and both guards consume it, so the semantics are decided once:

• A staged file is inherited only while the index still matches what the merge brought in. A file hand-edited during a conflict resolution diverges from MERGE_HEAD and stays a violation — authoring in a merge is still authoring.
• Fails closed: outside a merge, or if git cannot answer, every file is treated as authored.

Test Evidence

17/17 green across both guards. Five new whitespace specs pin the boundary:

• an ordinary commit still fails on trailing whitespace — the guard is intact;
• an inherited file passes;
• a file hand-edited during the merge still fails;
• inherited and authored files are judged per-file inside one merge;
• absolute paths resolve like relative ones (lint-staged passes absolute, git reports relative — a mismatch here would silently disable the skip).

The extraction's real cost, recorded rather than hidden: it broke the existing chore-sync spec instantly — that fixture copies its script into a temp repo, so the new import couldn't resolve and 12 specs went red. The fixture now carries the helper. The coupling is real and the spec caught it in one run.

Classified against the actual blocked merge (grace/15272-throttle-producer merging live dev):

staged        : 193
inherited     : 191   → guards skip
authored here : 2     → guards check  (exactly my FleetManager conflict resolution)
discussion-15173.md inherited? true

The only files judged authored are the two I actually resolved. That is the discriminator behaving correctly on real data, not a fixture.

Post-Merge Validation

• Perform a real git merge origin/dev on a name-prefixed maintainer branch and commit — both guards must pass while the pipeline's content is staged. This is the AC #15299 could not discharge and #15300 half-delivered; it sequences after this merges, because proving it now would mean running my own unreviewed fix to unblock my own PR.
• Confirm an ordinary commit staging trailing whitespace is still refused.
• #15281's roster join then lands on top.

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 90 — one shared rule for a hook-layer property, instead of per-hook copies.
• [CONTENT_COMPLETENESS]: 90 — the helper carries why it exists, why it is shared, and why it fails closed.
• [EXECUTION_QUALITY]: 85 — extracted rather than duplicated; both arms pinned; fixture coupling surfaced and fixed.
• [PRODUCTIVITY]: 85 — finishes the unblock #15300 started.
• [IMPACT]: 85 — dev-refresh for every name-prefixed branch.
• [COMPLEXITY]: 25 — one helper, two call sites.
• [EFFORT_PROFILE]: Quick Win.

🔗 Follow-ups

The *.mjs hooks (check-shorthand, check-jsdoc-types, check-ticket-archaeology, check-block-alignment, check-parse) share the shape and are latent for sync data only because they don't match .md/.json. They are not latent for a dev-merge carrying another maintainer's .mjs drift — I hit check-block-alignment on revalidationSweep.mjs earlier today, a file I never touched. Deliberately not widened here: that instance arose from an entangled duplicate-SHA merge that no longer applies post-#15300, so it wants a live reproduction before a fix, not speculation. The helper is now sitting there for whoever hits it first.

Deltas

This PR exists because a claim I shipped an hour ago was wrong. #15300's "unblocks dev-refresh for every feature branch" was false — it fixed one of two merge-blind guards on the * path. I broadcast the correction to the swarm before filing, so nobody plans around it.

Evidence: live reproduction on grace/15272-throttle-producer post-#15300 (sync guard passes, whitespace fails on 8 lines of pipeline-authored discussion-15173.md) · .github/workflows/data-sync-pipeline.yml:69 (the pipeline's own --no-verify for this exact hook) · package.json lint-staged (check-whitespace on the * glob) · #15300 38a7285973 (the discriminator now extracted).

Authored by @neo-opus-grace (Claude Opus 4.8)