**The gate could not see three of the shapes it exists to check** —
including the wiring I landed this week in #2990 and #3004. Found by
using it: the baseline still listed `useBlockerFanout.ts`,
`ExecutorStatusBar.tsx`, and `TaskDetailModal.tsx` as unwired *after*
those PRs merged.
### 1. Conditional shapes
Passing lanes only when they resolved is the **correct** way to write
these — an empty trait index means "not loaded yet", not "nothing is
terminal", so the caller must fall through to the documented legacy
default rather than fabricate one. Both idioms that produces were
invisible:
```ts
computeBlockerFanoutMap(tasks, flags ? { columnFlagsByTaskId: flags } : {}) // ConditionalExpression
computeBlockerFanoutMapCore(tasks, N, { ...(flags ? { classify } : {}) }) // SpreadAssignment, name === undefined
```
Either branch supplying the lane now counts. **Neither branch supplying
it is still unwired** — that negative is tested.
### 2. Vocabulary
`columnFlagsByTaskId`, the per-task trait index the dashboard threads,
was never added. It answers every lane question at once, so a call site
dropping it reverts to the legacy vocabulary wholesale — and the gate
would have stayed silent.
### 3. Name collisions — the false positive
Declarations were `set` by name, so the **last one parsed won**. Core's
`computeBlockerFanoutMap(tasks, n, opts)` and the dashboard wrapper
`computeBlockerFanoutMap(tasks, opts)` put their lane options at
**different argument indices**, so core's callers were checked against
the wrapper's signature: `task-priority.ts:141` passes `terminalColumns`
at index 2 and was reported unwired.
I caught this because adding the vocabulary entry in (2) made it appear.
A ratchet that reports a correctly-wired site is worse than one that
misses it — the first person to open one learns the number is noise.
Both shapes are now merged; a call satisfying either counts.
### Measured
| | |
|---|---|
| new tests | **6** — every positive paired with its negative |
| against the old census | **3 of 6 fail** — exactly the three
positives; the negatives pass either way, which is why they exist |
| baseline | **23 → 19** — four sites recognized as *already* wired; no
site newly excused |
| new flags | none |
### Verification
`node --test scripts/__tests__/check-lane-wiring.test.mjs` **16 passed**
· `pnpm test:gate` 161 + 13 + 487 + 71 · lint · lifecycle census
`--strict` · fnxc-dates (TZ=UTC) · changesets — green.
Seven other `scripts/__tests__` files fail on main independently of this
change (`verify-fast`, `dependency-security-floor`,
`engine-vitest-gate-policy`, `plugin-authoring-docs`,
`release-prompt-gate`, `ci-test-shard-timings`,
`workflow-reliability-release-check`). None are in the merge gate and
none are touched here — noting them because I looked, not because this
PR affects them.
ffe9898710