Claiming `packages/engine/src/scheduler.ts` from the census work order.
## Census before/after
| File | Before | After |
|---|---:|---:|
| `packages/engine/src/scheduler.ts` | 12 | **2** |
Measured with `scripts/lifecycle-column-census.mjs` (kind `column`
only).
## The file already had the right shape — it just under-answered
`resolveTaskParkedColumnsSync` already resolves a task's lanes from its
own workflow, **synchronously on purpose**: these run inside
`task:moved` / `task:updated` listeners, and its own comment records why
an `await` is forbidden there — it would defer everything after it to a
microtask and reorder handlers relative to a synchronous emitter. It
also already fails soft to the legacy ids.
But it only returned `{hold, intake}`, so every *other* lane question in
the same listeners was still asked with a literal. Widening it to the
full role set converted ten sites with no new abstraction, no new
resolution per site, and no change to the event-ordering contract.
## What was silently broken on a renamed board
- **PR monitoring never started** (`to === "in-review"`) and **never
stopped** (`from === "in-review"`) — a card's PR either untracked, or
tracked forever with its buffered comments never drained.
- **Terminal cleanup never ran** (`to === "done" || "archived"`).
- **The wip → hold failure bookkeeping never recorded** (`from ===
"in-progress"`).
None of these throw. They just stop happening — which is why the census,
not a red test, is what found them.
## Remaining 2, deliberately not converted
`L1097` (`task.column === "in-progress"`) and `L1171` (`task.column !==
"in-review"`) sit outside the listener where `parked` is in scope. They
need their own resolution, and resolving per call there is a different
cost profile than one-per-event; I flagged rather than guessed, per the
fleet rule.
## Verification
- census: `scheduler.ts` 12 → 2
- `scheduler-workflow-cutover` + `scheduler` — 42 tests green
- `tsc --noEmit` on `@fusion/engine` clean; `pnpm lint` clean
Behaviour on an unresolvable workflow is unchanged: the widened helper
keeps the same fail-soft legacy defaults the narrow one had.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
24f5ffaffa