[ben-vargas]

Fixes the intermittent viewer failure seen on run 37967628995 (attempt 1, viewer suite · node 18.17.1, PR #7 — which touches no viewer code):

FAIL src/app/App.test.tsx > two roving lists, one keyboard (§2.7 × §3.6) > moves only the rail on the rail's j/k — Home's cursor stays where it was
AssertionError: expected '#/run/r_2f91c4a8' to be '#/run/r_a03d51e7'

The race

mountShell() waits for the rail's rows. The test then sends j to Home, and nothing waited for Home to be ready:

• Home's table is a separate poll from the rail's, so it can commit a beat later.
• Even after it commits, Home answers keys through a window listener that React re-subscribes in a passive effect, which runs after the commit that painted the rows.

A j that lands in either gap runs against runs = [], so Home's cursor stays on row 0, and the final Enter opens RUNS[0] (r_2f91c4a8). That's CI's exact failure.

The fix

One test-only change. Before pressing j, the test waits for Home's listener itself:

await waitFor(() => expect(fireEvent.keyDown(window, { key: 'Home' })).toBe(false))

Home claims Home (calls preventDefault) only when it holds rows. No other window listener handles that key. Home also leaves the cursor on row 0, where it starts anyway. All of the test's assertions are unchanged. The table is virtualized in the shell, so Home's cursor can't be read from the DOM there, and waiting for Home's rows to render wouldn't close the second gap.

Verification

	Node 18.17.1	Node 24
Old test, Home's listing held 30 ms behind the rail's	25/25 fail (CI's message when Home's rows land between j and Enter, #/ when they land after Enter)	25/25 fail
Fixed test, same 30 ms lag	25/25 pass	25/25 pass
Fixed test, unmodified, repeated	50/50 pass (also 50/50 under 20 CPU hogs)	50/50 pass
Full viewer suite (vitest run, minus the local-only e2e) + typecheck	green	—

• Delaying React's passive effects (Scheduler host callbacks), with no fetch lag, also reproduces CI's exact message. That confirms the second gap is real on its own.
• The test still guards the bug it was written for. I re-broke the original hijack: removed the rail's stopPropagation, and Home's defaultPrevented and ownsShortcut checks. Both the old and the fixed test fail with the same message.

No CHANGELOG entry, since this is a test-only fix (same as the two test fixes in #8).

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.