bugfix: artifact-canvas test 'surfaces a synchronous FileAdapter.watch() failure via onError without throwing (D2)' flakes under CI load (missing waitFor on DOM assertion)

[amrmelsayed]

Problem

The test surfaces a synchronous FileAdapter.watch() failure via onError without throwing (D2) in packages/artifact-canvas/src/components/__tests__/artifact-canvas.test.tsx flakes under CI load. It passes 5/5 isolated and 8/8 full-file locally; fails intermittently in CI. Surfaced by the AIR builder for #1108 (PR #1109), which observed the failure on a VS Code-only diff with no artifact-canvas changes (git diff main...HEAD -- packages/artifact-canvas empty on the builder branch).

Root cause (verified against source)

The test waits for onError but synchronously asserts the DOM in the next line:

// packages/artifact-canvas/src/components/__tests__/artifact-canvas.test.tsx
await waitFor(() => expect(onError).toHaveBeenCalled());
expect(document.querySelector('p[data-line]')).not.toBeNull(); // read succeeded → content still renders

The waitFor ensures onError fired, but the very next assertion is a synchronous DOM read with no second waitFor. The async read() render in ArtifactCanvas can resolve AFTER onError fires — under CI load the render is delayed enough that the DOM query lands before the paragraph is in the tree. Race; test fails.

The sibling test immediately below (Disposable.dispose is safe to call more than once (idempotent contract, D2)) uses the correct pattern, which is the same fix to apply here:

await waitFor(() => expect(document.querySelector('p[data-line]')).not.toBeNull());

Fix

Wrap the second assertion in waitFor:

await waitFor(() => expect(onError).toHaveBeenCalled());
await waitFor(() => expect(document.querySelector('p[data-line]')).not.toBeNull());

One-line change. The semantic stays identical (the assertion eventually succeeds), only the timing tolerance widens to match the sibling test's pattern.

Verification

• Test must pass consistently when the existing test file is run under load (e.g., vitest --runInBand alongside other dashboard suites that compete for the event loop).
• Existing passing cases in the file must continue passing (no functional change beyond timing).
• Full pnpm --filter @cluesmith/codev-dashboard test suite stays green.

Out of scope

• Auditing the broader artifact-canvas test file for other similar await waitFor → bare-assertion patterns. Could ride along if the fixer notices any (the file is small), but not required.
• Any production code changes to ArtifactCanvas itself. The test is wrong, not the code.

Protocol

BUGFIX. Trivial, isolated, mechanical fix with a clear root cause; no design decisions, no architectural impact.

[amrmelsayed]

On it! Working on a fix now.