The largest remaining census cluster. Deferred twice, with a blocker
that turns out to be false **in the same component where its
counter-example already lives**.
## What the earlier notes got right
`detailColumnFlags` describes the **modal's own task**, and the column
classified here belongs to a **neighbour**. Supplying it would answer
*"is this neighbour finished?"* with a different row's traits — wrong on
data, not merely stale on vocabulary. That reasoning stands and I kept
it.
An earlier pass also converted this, left the parameter unsupplied, and
**reverted it** — correctly. An unsupplied optional parameter is
strictly worse than the literal: the guard is gone, the census counts a
conversion, and the behaviour is the legacy fallback forever. That rule
is why the wiring ships in this same commit.
## What the conclusion got wrong
> "A correct conversion needs per-**neighbour** flags — which the modal
does not have and should not fetch mid-render."
`columnFlagsByTaskId` is a per-task map. It is **already a prop** of
`TaskDetailModal` (declared :367, destructured :727), and the call site
at :992 sits **below** that destructure.
And `TaskDetailModal` already uses it exactly this way, for the
near-duplicate canonical:
```ts
columnFlagsByTaskId?.get(nearDuplicateCanonical.id)
```
…under a note observing that *its* blocker had been *"asserted from the
shape of the problem rather than tested against what was in scope."*
Same assertion, one function over. So the supplier the earlier note went
looking for exists, is per-neighbour, and needs no fetch.
## What it fixes
This lookup skips **finished** candidates so a done/archived prior undo
attempt never renders as an active "Undo task" link. On a board that
renames those lanes it matched neither — a finished undo task kept
rendering as open, which is precisely the stale affordance the
function's own header says it exists to prevent.
## Census
| | before | after |
|---|---|---|
| `taskRevert.ts` | 2 | **0** |
| repo backlog | 17 | **15** |
## Measured
- 4 new cases; `taskRevert.test.ts` **11/11 pass**.
- **MUTATION**: restoring the literal pair fails the renamed case.
- **The negative is load-bearing.** The map is fail-soft, so a candidate
it does not cover must still be treated as **open**, not skipped. A
conversion that skipped unknown candidates would *hide live undo links*
— failing in the direction nobody reports.
- A **control** pins that an unwired caller (no flags at all) still
skips the legacy ids, so the optional parameter cannot regress default
boards.
- `TaskDetailModal` suites — **31 files / 664 tests pass**.
- `tsc --noEmit -p tsconfig.app.json` clean; census `--strict`,
`check-lane-wiring`, `check-fnxc-future-dates` clean.
## Pattern worth noting
This is the fourth deferral this session whose stated blocker had
dissolved or misidentified itself, and the second where the
counter-example was already in the same file. The common shape: a note
records *why* something is blocked, is accurate when written, and is
never re-checked — so the block outlives its cause. Re-reading them cost
minutes each and returned two real conversions.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
a8dae03fdb