# Real-app evidence procedure

Both runs used Linux Chromium 157, 1280 x 800 CSS pixels, dark appearance.
Each commit ran its own web dev server and server from its own detached worktree.
Dependencies were installed with vp i. Each server was launched with
T3CODE_DEV_SEED=0 vp run dev. No live database was read or copied.
Each server created fresh state; no user content or provider turn was needed.

Base: 9cd7a04797184c92a8f636ae1e918f42c139d850
Head: d89e3ace618a1f10ab86304d460970e344b4e404

1. Open the actual dev app and pair using its startup token (excluded from uploads).
2. Inspect the browser accessibility tree with no toast present.
3. Open /settings/about; inspect the empty Notifications regions again.
4. Invoke captureSide(tools, tabId) from capture-driver.js on the base and then head.
   This is a T3 preview-tool orchestration script; tools is the preview-tool object.
   The same function was evaluated and invoked for both live browser tabs.
5. Compare the two PNGs returned by each invocation with cmp. Both pairs matched.
   The driver waits 500 ms between screenshot requests.
6. Read the captured accessibility tree excerpts and viewport attributes.
7. Observe the recorded click on Copy version and the visible Copied! confirmation.
   A read-only MutationObserver records additions to the polite live viewport.

The image annotation is outside the captured app, and quotes the runtime
accessibility tree excerpt. The original unannotated stills are also uploaded.
Recordings are unmodified captures of the actual app. Every selected recording
was reviewed with extracted frames, including the Copied! interval.

Observed: two empty Notifications regions on base; one on head.
Both runs insert Copied! into toast-viewport-anchored with aria-live=polite.
On head, that viewport has no region role or Notifications label.

NVDA was not available on Linux. This proves Chromium accessibility exposure
and live-region content insertion, not audible speech or NVDA keyboard navigation.
Electron was not separately exercised.

AI-assisted; verified with the tests and tools listed above.
