TR2 default OK activation and current controls — 2026-11-08

Native default-OK Enter works through production Worker control. The current module reaches Venice, and ordinary Up displacement and a separate Left heading change persist after release. No emulator mismatch was demonstrated or repaired. Evidence: scratch/runs/20261108-tr2-default-ok-activation/.

The preceding correction was followed: visible Graphics captures were reviewed as Graphics, with their raw hashes. The earlier startup/fresh/refresh identity does not establish disappearance or an emulator defect. Neither prior box was resumed or deleted.

Local qualification before allocation

One local browser used the unchanged registered original and production index.html?debug&threads&app=tomb_raider_2_demo&gl-renderer=webgl&d3d-renderer=webgl. local-default-ok/fresh-local.png, SHA-256 1e48ba49c6358f78ee24dfc3bda38c2510c243949d6bf248cb070645acb8d03a, was inspected at original resolution: native Graphics, Hardware 3D Acceleration, Full Screen 640×480 High Color, and the visible default OK outline. Ordinary trusted CDP Enter, held 100 ms with independent release, was admitted at age 7455 ms, inside the unchanged 12000 ms guard. The exact ACK and after-enter.png establish dismissal and textured Venice. The existing one-way native-to-producer binding succeeds with the actual guest and render modules. This qualifies the real production Worker route, beyond a mocked key delivery.

The local browser closed normally, exit0, at 06:05:22.832 UTC. Both controller and Chrome PIDs and its profile were absent before remote allocation. Local chain controls passed owner replacement, stale zero-effects, one-way binding and independent release under a slow sampler. Atomic transport controls passed exact ACK, duplicate zero-effect, stale-only-refresh and sticky UNKNOWN tests.

Fresh-box original and persistent controls

Published source was 78df2e4d90114320f68ed2b44014f6d10d929e06; the route, registered executable/media, configuration, clocks and seed were unchanged. No private input handler or guest memory write was used. The driver reused the qualified key path and one-way native-to-producer binding without weakening owner, module, image, freshness, ACK or UNKNOWN checks.

The unchanged demo proceeds directly from native OK into Venice; no NewGame screen was presented and no separate NewGame selection is claimed. There were eight exact successful original ACKs: four passive captures, native Enter, Up, Left and normal quit. No stale refusal, partial input, transport UNKNOWN, replay or second original occurred. Weapons, target consequences and audio were not tested here.

Identity and evidence checks

Both actual guest and owning render Worker modules are SHA-256 7ec49cbbae4cb177909fd8fd9d3599713ee5f2a5ffb7f7e8c890848a540052bf. Guest target 13F5EFAC26F8E52448AD7212E9D03494, endpoint 1, slot 0, transitions once to render target C65DA35EEEB1B0B7E2749B81402AC302, endpoint 1, {api:'legacy',backend:'webgl'}. The owning Worker context observer reports WebGL 2.0 and ANGLE (Intel, Mesa Intel(R) UHD Graphics 620 (KBL GT2), OpenGL 4.6 (Core Profile) Mesa 23.2.1). The adapter wrapper's direct context field is null; the context identity comes from the observed owning Worker, not that field. The final owner is ready with error0, failed=null and software completions0.

Twenty recovered-module checks, four loaded host-source hashes and 184 successful served response hashes match pins. All eleven local/remote PNGs pass chunk CRC and IDAT inflation; originals were reviewed without renaming Graphics as desktop. All 594 remote input pins remain unchanged after exit. Git SHA, relevant worktree diff, helper patches and existing media path/hash references preserve source identity. No source/media archive or copied media was placed in evidence, and no fresh-build claim is made.

No frame collector was qualified. Notification, queue, completion and publication counts are not logical or displayed FPS. Frame association, observer overhead, pacing windows, full27, backend parity, audio and unmeasured FPS remain open. The next measurement task must qualify frame identity before reporting rates; this run establishes current bounded movement and heading.

Drain, retrieval and normal stop

One UUID-idempotent box, bx_av37ex7j, was created at 06:05:48.045 UTC after independent normal/watchdog delegates were observably prearmed. Its immutable 25-minute deadline was 06:30:47.870 UTC; initial expiry 06:30:48.061 UTC. Only one heavy runtime ran on that box. Sampled local disk stayed above 5 GB. No RAM gate, old-box operation, nested worker or deployment occurred.

Chrome closed normally at 06:07:32.007 UTC, exit0/signal null. Two immediate read-only drain checks still saw the controller PID; those failures remain preserved in their remote results. The final 06:08:01.515 UTC check proves both PIDs absent, HTTP/CDP ports refusing, profile/writer absent, no matching processes and all input pins unchanged. No additional game input followed quit.

All 106 remote artifact references, 96 unique hashes, were retrieved and independently reread for size/SHA-256 at 06:08:14.134 UTC, before requesting normal stop. The final drain artifact supersedes the early failed check in the final inventory; the original inventory and both earlier results are retained. Box state is archived, active expiry null, stopped 06:08:15.972 UTC. Receipt stop_569b38f082eb35c2 completed 06:08:15.993 UTC, forced=false, error=null. Both delegates exited normally; supervisor children are empty.