Extracted from the Tier-2A run record #9417 (QA wave #9296), where it is the round's single clause failure — cli.migrate-meta-codemod (P1, rev 1), clauses 1 & 2. Subject sha e4e5c6e3c608b1b807c83a0d5b734f213eb1a1dd.
The defect
os migrate meta --from <N> loads the stack config through the current, post-retirement schema before any conversion runs. A source carrying a retired key therefore fails validation and the command exits 1 having rewritten nothing.
That input — a source with retired keys — is the only input the command exists to handle.
Why it matters more than a P1 usually does
The refusal it prints is the very instruction that sends people there. Run \os migrate meta --from ` to rewrite existing sources automaticallyappears **144 times across 39 files** underpackages/spec/src(verified independently of the run:git grep -c "os migrate meta --from" origin/main -- packages/spec/src`). Every one of those retirement messages promises an automatic fix that the tool cannot perform on the input it was told to fix.
So the user-visible loop is: author hits a retired key → spec tells them to run the codemod → codemod refuses because of the retired key → spec tells them to run the codemod.
Evidence and controls (from #9417, reproduced twice)
- The retired key is the source's only validation issue — nothing else in the file is invalid.
- Same refusal at
--from 17 and --from 3, so the ADR-0087 floor check is not the gate either; the failure is upstream of version dispatch.
--out is not implicated.- The command behaves normally on a clean tree, so it is not broken generally — only on the input class it targets.
Not yet probed
The --stored arm (replaying over sys_metadata rows) is a different code path and was not exercised. Worth a probe before the fix is scoped: if it shares the pre-validation step it has the same defect; if it does not, the fix may only need to bring the file arm in line with it.
Fix direction (not prescribed)
The conversion must run before the source is validated against the current schema, or the load must be tolerant of keys the target version retired. Whatever lands should be pinned by a test whose fixture is exactly the failing case — a source carrying a retired key, migrated end-to-end — since a test on a clean source passes today and proves nothing.
Left unlabelled for domain:*: routing is the triage seat's call (the fix lands in packages/cli, possibly with packages/spec retirement-message coordination). Priority likewise — filed P1 by its checklist item, but the 144 messages prescribing it are an argument for re-levelling that belongs to triage, not to the runner or the wave PM.
Extracted from the Tier-2A run record #9417 (QA wave #9296), where it is the round's single clause failure —
cli.migrate-meta-codemod(P1, rev 1), clauses 1 & 2. Subject shae4e5c6e3c608b1b807c83a0d5b734f213eb1a1dd.The defect
os migrate meta --from <N>loads the stack config through the current, post-retirement schema before any conversion runs. A source carrying a retired key therefore fails validation and the command exits 1 having rewritten nothing.That input — a source with retired keys — is the only input the command exists to handle.
Why it matters more than a P1 usually does
The refusal it prints is the very instruction that sends people there.
Run \os migrate meta --from ` to rewrite existing sources automaticallyappears **144 times across 39 files** underpackages/spec/src(verified independently of the run:git grep -c "os migrate meta --from" origin/main -- packages/spec/src`). Every one of those retirement messages promises an automatic fix that the tool cannot perform on the input it was told to fix.So the user-visible loop is: author hits a retired key → spec tells them to run the codemod → codemod refuses because of the retired key → spec tells them to run the codemod.
Evidence and controls (from #9417, reproduced twice)
--from 17and--from 3, so the ADR-0087 floor check is not the gate either; the failure is upstream of version dispatch.--outis not implicated.Not yet probed
The
--storedarm (replaying oversys_metadatarows) is a different code path and was not exercised. Worth a probe before the fix is scoped: if it shares the pre-validation step it has the same defect; if it does not, the fix may only need to bring the file arm in line with it.Fix direction (not prescribed)
The conversion must run before the source is validated against the current schema, or the load must be tolerant of keys the target version retired. Whatever lands should be pinned by a test whose fixture is exactly the failing case — a source carrying a retired key, migrated end-to-end — since a test on a clean source passes today and proves nothing.
Left unlabelled for
domain:*: routing is the triage seat's call (the fix lands inpackages/cli, possibly withpackages/specretirement-message coordination). Priority likewise — filed P1 by its checklist item, but the 144 messages prescribing it are an argument for re-levelling that belongs to triage, not to the runner or the wave PM.