**`pnpm test` and `pnpm test:full` fail on `main` right now** (commit
`51934931e1`). Found by running the gates against `main` after my
earlier PRs landed, not by CI telling me.
> **Correction to this PR's first version.** I originally wrote that
`pnpm test:gate` was failing. It is not: `check-inert-flag-seams` runs
in the `pretest` and `pretest:full` hooks, not in `test:gate`. So the
blocking merge gate (Lint / Typecheck / Build / Gate) is unaffected and
PRs are not blocked — what is broken is every local `pnpm test` run,
which fails before a single test executes. Lower urgency than I claimed,
still worth fixing promptly, and I would rather correct the scope than
leave an overstated one standing.
Three causes, each surfaced by fixing the one before it.
## 1. Stale exemptions — the mechanism working
#2819 and #2823 merged, so the two `ALLOWED_OMISSIONS` entries covering
those call sites became stale and the staleness check failed them.
Removed. This is my cleanup: the entries were designed so they could not
outlive their fixes, and they didn t.
## 2. Renamed imports were not resolved
Removing the first entry surfaced:
```
enqueueMergeQueue() — best call passes 2 of 5
```
Its only production caller passes all five — through `import {
enqueueMergeQueue as enqueueMergeQueueAsync }`. Call sites were recorded
under the **local** name, so an aliased supplier was invisible and the
seam read as unsupplied. The local name is now mapped back to the
exported one.
## 3. Method calls were conflated with module functions
That fix then surfaced two engine sites as omitting — but
`store.enqueueMergeQueue(taskId, opts)` is a **2-arg `TaskStore`
method** that resolves the review columns internally (#2819), not the
5-arg module function sharing its name. Property-access calls are no
longer attributed to module-level seams.
**Tradeoff, stated at the site:** a genuine `namespace.fn(...)` call is
now skipped. This codebase calls module functions as bare identifiers,
and aliases are resolved by fix 2, so that shape does not currently
occur. Recorded rather than left for someone to discover.
## Both directions re-verified
A fix that quietly disarms the gate would be worse than the red, so I
re-ran the defects it exists to catch:
| mutation | result |
|---|---|
| drop `Column.tsx`'s flags argument (partial supply) | `supplied by
10/11 call sites; omitted at .../Column.tsx:1 (of 2)` |
| drop the aliased 5-arg supplier (wholly unsupplied) | `best call
passes 3 of 5` |
| restored | exit 0 |
My first attempt at the second row grepped for the wrong message shape
and printed nothing. **I re-ran it rather than reading silence as
success** — which is the failure this gate exists to prevent, and one I
have made in this same file before.
## Verification
`pnpm test:gate` green (it was never affected) · `node
scripts/check-inert-flag-seams.mjs` exit 0 · lint 0 · gate reports `21
lane/flag seams, all supplied at every production call site`.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
5a3541315c