## Root cause
`Board.tsx:874` renders `BoardWorkflowSkeleton` whenever `boardWorkflows
=== null || boardWorkflows.workflows.length === 0`.
All three files depended on the **legacy board**, which U12/R9 deleted —
along with the `flagEnabled` conjunct in that condition. So
`flagEnabled: false` now selects nothing, and the board sits at
`"Loading workflow lanes"` with no lanes, no task cards, and no
Auto-merge toggle.
Every failure surfaced as `Unable to find an accessible element with the
role "checkbox" and name "Auto-merge"` — which points at the query, not
at the board state. The `aria-label` in the error's role dump is what
gave it away.
## Three causes, in sequence — each only visible after fixing the one
before
| # | File | Cause |
|---|---|---|
| 1 | `auto-merge-toggle-blank.mobile` | mocked `workflows: []` →
empty-lane skeleton |
| 2 | `auto-merge-toggle-blank.mobile-integration` | builds its mock
with `createDashboardApiMock`, which **spreads the real module** — so
`fetchBoardWorkflows` was the real network call, never resolving under
jsdom |
| 3 | both | even with a payload, lanes arrive on a **promise** while
these synchronous tests assert immediately after `render()` |
(3) is the one worth remembering: the legacy board rendered
**synchronously**, so no await was ever needed. `renderBoardWithLanes`
flushes a microtask inside `act` — a microtask and not a timer advance,
because this suite runs on fake timers.
## Also re-pinned: a source-scan guard pinning deleted code
`board-mobile` asserted `ref={setBoardRef}` appears **3** times across
"legacy, selected, and aggregate" renders. The legacy `<main
className="board" id="board" ref={setBoardRef}>` render has **0**
occurrences now, so both the count *and* its `toContain` were pinning
removed markup — the second would have failed as soon as the first was
fixed.
Now **2**, with the legacy markup asserted **absent** so it stays gone,
and the test renamed to the two renders that exist. The guard's real
point — that both share one scroll-snap hook — still holds.
## Measured
| Check | Result |
|---|---|
| `components-a` group | 22 failed → **0** (51 files, **1195 passed**) |
| lane payload emptied again | **all 8** cases in the first file fail —
load-bearing |
| `pnpm lint`, dashboard app `tsc` | clean |
## One wrong turn, recorded
My first attempt declared the payload as a module `const` and referenced
it from the `vi.mock` factory. Factories are **hoisted above const
declarations**, and the file reported `"no tests"` rather than a
hoisting error — a failure mode that reads as a collection problem, not
a reference problem. Inlined instead.
## Dashboard status
With #2735 (50 → 1) and this (22 → 0), dashboard goes from 88 failures
across 9 files to **17 across 6**. The remainder are unrelated: CSS
`toHaveClass` / computed-colour assertions, a missing
`wf-add-step-modal` testid, and the one GitHub-tracking affordance
question flagged in #2735.
Per #2732 these lanes are currently **never executed in CI** — the shard
aborts on the first failing package — so this is cleared ahead of them
starting to run.
a09263ae8c