Owner TR2 WebGL release check — 2026-11-07

TR2 renders the Venice scene through its actual WebGL Worker, but this attempt does not qualify ordinary gameplay, sustained rate, or release suitability. A fresh-capture review error ended the single original. No emulator defect was demonstrated; no implementation or safety guard changed.

Evidence: scratch/runs/20261107-owner-tr2-gl-rate/. Owner Telegram172 requested GL checks beginning with TR2. GTA2 and TR3 were not launched in this task.

Terminal blocker

The production direct URL was index.html?debug&threads&d3d-renderer=webgl&app=tomb_raider_2_demo. The original began at 02:53:05 UTC. The personally reviewed initial capture showed Graphics, Hardware 3D Acceleration, Full Screen 640×480 High Color and Z Buffer. There was no render endpoint yet.

The first ordinary OK request was refused before dispatch: stale scene, effects=0. A separate capture completed at 02:53:40 UTC. That fresh review/options-fresh.png shows only the teal desktop, without the Graphics sheet or OK. The operator nevertheless submitted an ordinary click with an incorrect inherited description saying that Graphics and OK were visible. This mismatch invalidates the input evidence. The original records remain unchanged and capture-reviews.json explicitly corrects the description. Whether the missing dialog in that capture reflects transient composition or another cause is unknown; it is not an established emulator defect.

The complete click ACK produced retrieved/original/venice-idle.png, personally reviewed after stopping input: Lara stands idle in the textured Venice alley, with the slope and orange lattice visible. This establishes scene rendering. No forward, turn, weapon toggle, combat or further gameplay input was sent. No controlled stepping, second original or retry followed the blocker.

All four original queue commands have complete ACKs: one zero-effect stale refusal, one refresh capture, one balanced click, and ordinary quit. There was no partial or UNKNOWN action outcome and no ID replay. A complete ACK does not repair the false review description or establish gameplay quality.

Actual renderer and rate limits

Direct CDP inspection identifies render target C41962E58582DE9864B63032CA5B2722, endpoint 1, legacy/webgl, adapter present, failed=null, ready=1, error=0, completed sequence=140, batches=93 and commands=7816. At the Venice capture, naturally returned owning-endpoint counters report 7439 draw calls, 9863 triangles, errors=0, fallbacks=0, software completions=0 and fallback-software completions=0. These samples are sequential, not atomic.

The browser GL renderer string from that endpoint is:

ANGLE (Intel, Mesa Intel(R) UHD Graphics 620 (KBL GT2), OpenGL 4.6 (Core Profile) Mesa 23.2.1)

Version: WebGL 2.0 (OpenGL ES 3.0 Chromium); Chrome 151.0.7922.108. This reports the owning context, rather than inferring acceleration from a toolbar. No separate GL-call instrumentation or physical-display measurement was installed. Adapter statistics are closure-private; the generic direct adapterStats field is null, while natural endpoint responses supply counters.

Wall-clock sustained presentation rate, logical update rate and frame-time distribution are unknown. The attempt ended before a qualified sustained sample. Draw totals and 21 reported transport flips are not FPS. TR2 has no registered logical-frame marker in the staged app entry. No rate is calculated from launch duration, scheduler slices, browser rAF or these cumulative totals. The idle capture alone does not establish rendering parity or release quality.

Qualification and identities

Before the original, nine atomic transport cases, seven actual executor/recovery cases, seven production direct-initializer cases, production-renderer physical key state and invalid-duration controls passed. Native CDP trusted balanced Up/Left/Space/Enter/Escape pairs passed on a generated blank page. The unchanged scene guard requires a current reviewed capture within 12 seconds; holds remain 50–2000 ms, with invalid durations rejected before effects.

On this fresh boat, production direct launch of Boids established a real guest Worker and actual healthy legacy WebGL owner. Its naturally returned draw-call counter advanced 335→469 and direct completed sequence advanced 40→49, with fallbacks=0 and the same Intel renderer string. These are qualification counters, not a Boids performance measurement. A separate generated PE loaded pinned native COMCTL32, DllMain returned EAX=1, and the personally reviewed ordinary tabbed property sheet dismissed with a balanced OK click. Both qualification sessions were fully cleaned before the original.

Staged source commit: 0daaddf62c8013e57fdb94756d630f4acda349d7. Relevant dirty/new-source patches and owned helper patches are retained. Git remains the source archive. Both actual original Workers yielded Wasm SHA-256 07f1cc970038f8ab31313b8b200edf26caa63473f293f5fc937f5cfe5c5c7f6e (2,181,179 bytes). This existing module was reused; staging HEAD is not asserted to be its fresh build provenance. All 23 executing JavaScript sources from two original Workers matched staged file bytes.

All 626 serving/helper pins match remotely before the original and after exit; all 172 successful original HTTP responses match pinned bytes. Non-200 responses remain in http-audit.json. Fixture identity matches retained TR2 evidence, including the following existing files under test/binaries/:

File SHA-256
candidates/tomb-raider-2-demo/game/TOMB2.EXE 0811efa84bfab8c458b9e29976088152db4c14601a99ca4910c75a2d009233aa
candidates/tomb-raider-2-demo/game/DATA/DEMOPC.DAT e1bdd5776e1946420e0eeb9e1a4da12f16ea2bbfb7c3909b414372b100ff51bd
candidates/tomb-raider-2-demo/game/DATA/BOAT.TR2 e710716fcaf039fb3462e26e33f0ead8eb45bd01c8bbfa036c455a1d19823d9e
dlls/comctl32.dll 28a41e6a35c4509924a32a99841a815cf49e1d3ee6912ec26a823e45472edfa4

No media/config/save file bytes were edited. Upstream provenance and historical native reference evidence remain unavailable; local hash agreement does not recover them. Individual serving inputs were transferred without source/media archives. No local browser/emulator/source/build lane was used; only offline helper controls and identity preparation ran locally. COM ABI qualification was compiled and tested on the separate boat.

Retrieval and completed stop

All 122 inventoried remote run files were retrieved or matched to retained identical bytes; module duplicates share a retained file with explicit mapping. The original Chrome exited normally, code 0, signal null, at 02:54:07.692 UTC. The first cleanup sample caught its controller finishing exit; the independent 02:54:33 check confirms all six tracked controller/browser PIDs absent. All six ports refuse connections; all three profiles and writer locks are absent. The cleanup template's pending:'UNKNOWN; never replay' is generic wording, not an UNKNOWN command outcome.

Fresh boat bx_6fupqmmb used a persisted, fsynced UUID Idempotency-Key. Explicitly delegated normal stop stop_c438906cecc10c99 completed at 02:54:38.459 UTC, forced=false, error=null. Box state is archived, active expiry null; initial expiry was 03:50:38.331 UTC. Allocation through completed stop took about four minutes, within 35 minutes. Observed local disk remained above 2 GB. No automated risk flag, held-risk retry, old403 resume, nested worker or public deployment occurred.

The next authorized TR2 attempt needs accurate fresh-scene review before every ordinary action and a separately qualified sustained collector with explicit counter semantics. This run supplies no release recommendation. Publication, corrections and validation precede result.json, written last, then stop.