Canary probe asserts against markup two PRs replaced

[mforce]

The canary probe (tools/simulation/ui/specs-canary/canary.spec.ts) asserts against SPA markup that two merged PRs replaced. Two of its four screens can no longer pass, and nothing noticed because the probe is workflow_dispatch-only — the pull_request run of e2e-smoke.yml skips the canary step.

Evidence: run 34787979138, on the same head SHA (eabae19) whose pull_request run was green.

1. dashboard — the populated-rows assertion is structurally unsatisfiable.

Error: dashboard rendered its table with no rows against a populated fixture
Locator:  locator('.capture-tile').first().locator('tbody tr')
Expected: not 0
Received: 0
Timeout:  60000ms

SCREENS[0].ready is .capture-tile, and line 154 asserts ready.locator("tbody tr") is non-empty. Since the dashboard rework (396ba23, #654) a .capture-tile is a <Link> holding three <div>s, and the dashboard has no <table> on it at all — tiles, a trend chart, a stock <ul> and a sales <ul>. That commit re-pointed ready at the new markup and left the row assertion behind, so the count is always 0.

2. history — the flock filter is no longer a <select>.

Error: no <option> containing "Sim House A" — the list did not load, or the label format changed
Locator:  getByLabel('Flock').locator('option').filter({ hasText: 'Sim House A' })
Expected: 1
Received: 0

canary.spec.ts:120 calls selectOptionContaining, which needs <select><option>. Since 60d2053 (#642, searchable entity pickers) HistoryPage.tsx renders a FlockPicker, so there are no <option> elements. specs/manager.spec.ts:129 already drives this same control correctly through commitNamedPicker.

The backend was healthy throughout: every request in the canary window returned 200 (/api/v1/flocks, /daily-entries, /stock, /reports/*), and the stock and reports screens passed. The 60s waits are timeouts on locators that resolve to zero elements.

Scope

• Give each screen its own populated-rows locator instead of the hardcoded ready.locator("tbody tr"); the dashboard's rows are its capture tiles.
• Drive the history flock filter through commitNamedPicker.
• Verify by running bash tools/simulation/ui/run-canary.sh against a seeded sim stack.

The drift itself is the deeper finding: the canary is dispatch-only, so a markup change that breaks it produces a green PR and a probe nobody can trust when they next reach for it.