Browser regression suite fails under pytest fixture (works standalone)

[techmore]

Context

While fixing CI (PR #229), the browser regression suite was found to fail consistently — both in CI and locally — even though each test's scenario works when reproduced standalone against a live server.

Symptoms

• All 7 tests time out waiting for socket-driven DOM updates (Locator.wait_for: Timeout)
• Server logs show repeated POST /socket.io/... 400 and websocket ... 400 for valid sids after initial connect
• The same flows work in a standalone headless-Chromium reproduction with zero page errors

Suspected cause

Interplay between the module-scoped browser_server fixture, socketio.test_client() sessions, and the page's socket session — possibly the shared broadcaster/client-state registry treating reconnects or the second client as stale.

Also note

~16 runtime-contract tests assert frontend content that only exists on the swift-native branch; those need contract reconciliation once the Swift line merges.

Acceptance

• Browser regression suite green in CI
• Runtime-contract suite reconciled with the merged frontend

[techmore]

Progress (2026-08-22)

Fixed on main (4224065):

• Fixture isolation: scan_fixture now mirrors entries into the runtime sqlite store - /api/runtime/reports reads only from the store, so JSON-index-only writes were invisible to the running server. This was the root cause for the reports/history tests.
• Tab-wiring race: tab switching is now wired on DOMContentLoaded before socket connect, so clicks no longer hit hidden panels.
• Strict-mode selectors: .first guards and history assertions scoped to the fixture's card (the shared data/scans dir accumulates entries across runs).

Result: 2 of 7 tests pass (reports tab render, history tab diff summary).

Remaining: protocol bridge

The other 5 tests (second-tab replay x4, live fanout) need a protocol bridge: the Flask runtime emits client_state_snapshot / auto_scan_status / job_status, while the Node-era frontend listens for sync_state / initial_data. The replay flows depend on the frontend applying the Flask-side events.

Suggested approach: emit legacy-compatible events alongside the new contract (or add frontend listeners for the new events), then re-enable the replay tests.