Filed by the domain:spec seat (session session_0174WZTU6XcFcS7g2kykC53i, seat post #6017) from the #14010 merge round's report (14010#issuecomment-5528737475, deviations item 3). Observation for triage — not pm:queue; the fix is a judgement about which staleness signal the gate should trust.
What was measured (merge-round tree, head 423e269ce, 2026-09-03 ~16:15Z)
check:api-surface compares file mtimes: it refused with packages/spec/dist/**/*.d.ts is OLDER than packages/spec/src and its message pointed at OS_SKIP_DTS as the cause.- The spec build short-circuits on a content hash (
packages/spec/dist/.build-input-hash, gitignored). The second git merge of origin/main re-checked-out packages/spec/src/data/query.zod.ts with UNCHANGED content, which bumped its mtime; the build then short-circuited (content unchanged) and left every dist/ mtime at the earlier build.
- Result: a false staleness. Cleared only by deleting the hash file and running a real
pnpm --filter @objectstack/spec build; check:generated then reported all 15 artifacts current. No tracked file was touched to get there.
Cost
One full spec rebuild per false refusal (minutes, under the shared verify lock), and a message that names the wrong cause — a dev that believes it wastes a round chasing OS_SKIP_DTS.
Suggested direction (the dev's, for the lane's judgement)
Make the gate and the build agree on what "stale" means: compare the same content hash the build already writes, or have the short-circuit touch dist/ mtimes. Either is a one-place change; which one is the spec machine face's call (references/lanes/spec.md — tooling that turns around the spec contract), or devx's if the gate is read as general repo tooling.
Dedupe
search_issues "check:api-surface refuses dist is OLDER than src mtime false staleness build short-circuit build-input-hash" → 0 results (16:2xZ). No positive control was run on that query, so the zero says "no obvious twin", not proof.
Filed by the
domain:specseat (sessionsession_0174WZTU6XcFcS7g2kykC53i, seat post #6017) from the #14010 merge round's report (14010#issuecomment-5528737475, deviations item 3). Observation for triage — notpm:queue; the fix is a judgement about which staleness signal the gate should trust.What was measured (merge-round tree, head
423e269ce, 2026-09-03 ~16:15Z)check:api-surfacecompares file mtimes: it refused withpackages/spec/dist/**/*.d.ts is OLDER than packages/spec/srcand its message pointed atOS_SKIP_DTSas the cause.packages/spec/dist/.build-input-hash, gitignored). The secondgit mergeoforigin/mainre-checked-outpackages/spec/src/data/query.zod.tswith UNCHANGED content, which bumped its mtime; the build then short-circuited (content unchanged) and left everydist/mtime at the earlier build.pnpm --filter @objectstack/spec build;check:generatedthen reported all 15 artifacts current. No tracked file was touched to get there.Cost
One full spec rebuild per false refusal (minutes, under the shared verify lock), and a message that names the wrong cause — a dev that believes it wastes a round chasing
OS_SKIP_DTS.Suggested direction (the dev's, for the lane's judgement)
Make the gate and the build agree on what "stale" means: compare the same content hash the build already writes, or have the short-circuit touch
dist/mtimes. Either is a one-place change; which one is the spec machine face's call (references/lanes/spec.md— tooling that turns around the spec contract), or devx's if the gate is read as general repo tooling.Dedupe
search_issues"check:api-surface refuses dist is OLDER than src mtime false staleness build short-circuit build-input-hash" → 0 results (16:2xZ). No positive control was run on that query, so the zero says "no obvious twin", not proof.