# Fixture matrix clarification for the final retake

This amends the trigger/seed details of the [pass-1 matrix](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/d4ce8c1f40d6375e/proof-audit-pass1.md) after the [supplemental pass-2 audit](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/2b7d3a9eabf2e597/proof-audit-pass2.md). The sixteen logical before/after pairs, two Settings consumers, viewports/themes, and observable behavior requirements remain.

Final head [8fbfc33c1a](https://github.com/yordis/t3code/commit/8fbfc33c1ac503048ca2983564318bc29143086a) has the identical source tree to the code-reviewed f871636989 revision. Before remains [dd5b5401e1](https://git.cafe/pingdotgg/t3code/commit/dd5b5401e1f240f4314af29edf431369d182f386).

For R2/R4, the lead accepts **explicit runtime-triggered stale revalidation** as the fixture trigger: load the real admin session normally, retain its browser registry and prepared environment, age it past thirty seconds, then call the existing appAtomRegistry.refresh(environmentSession.sessionStateAtom(realEnvironmentId)). This invokes the real loader and real HTTP endpoint; it must not set an atom, replace its result, mock grants or fabricate a response. Label it precisely; do not describe it as navigation-triggered remount. This is capture seeding, not a product/API change or a claim that a new refresh control exists. Observe Success before the trigger, waiting with previous Success during the retry, then actual success or Failure. No new production code or mirrored tests are required.

For R1/R3, isolate healthy app bootstrap from the measured slow environment. A separate synthetic server connected as a saved remote environment is an acceptable real-app fixture. Delay only that environment's actual session request; do not rely on ordinal first-N bootstrap bypass. A cold episode must observe no previous Success; a warm one must observe actual cached Success. Read-only state/request instrumentation may log tags, timestamps and boolean grant checks, never credentials or response bodies containing secrets.

The Browser panel repeatedly produced 1280x800/DPR2 files for requested 1440x900 output, and its screenshot sampling could not satisfy continuous-recording timing. The authorized fallback is agent-browser with isolated named sessions, native DPR1 screenshots, real continuous recording, reduced-motion emulation and explicit /tmp output. No T3 preview recorder or live T3-home access is permitted.

A pilot must prove geometry, correct request ownership, mounted Settings across the first timeout, and real cold/warm state identity before batching. Original supplemental captures do not satisfy any final acceptance row. Native mobile remains explicitly unverified under the dispatcher's exception. The final proof auditor must inspect the new artifacts and reproduce a capture before issuing SUFFICIENT.