Owner GTA2 WebGL and ordinary-rate check — 2026-11-07

The original GTA2 Wild Demo renders through its actual WebGL Worker and supports personally reviewed settled walking and native Enter vehicle entry. This run does not qualify sustained driving or steering, and supplies no release-quality recommendation. The ordinary walking sample was visibly slow: 77 successful screen upload/composition completions in 30.003 seconds, about 2.57 per second. These are not unique game frames or guest updates. No emulator defect was demonstrated and no production implementation changed.

Evidence: scratch/runs/20261107-owner-gta2-gl-rate/. This is the single-task follow-through to owner Telegram172, after reading the software GTA2 and corrected TR2/TR3 GL findings. Exactly one original launched, at 03:09:41 UTC, through index.html?debug&threads&d3d-renderer=webgl&app=gta2_demo. The production direct initializer selected the registered app while the dropdown remained Notepad. Miles used its original app-local /binaries/candidates/gta2-demo/installed/Program_Executable_Files/mss32.dll route. No catalog substitution or original media/config/save edit occurred.

Personally reviewed gameplay and terminal limitation

Every used capture was freshly retrieved and viewed with view_image, detail: original, before its geometry or scene description informed another action. Capture hashes, sizes and dimensions are retained in capture-identities.json; action-specific descriptions remain in request/review records. No inherited TR2 review or ambiguous TR3 display assessment was reused.

The terminal limitation is unresolved visual driving/steering qualification, not a transport failure or demonstrated emulator defect. No further gameplay input followed reverse1; executing identities were retrieved and the browser closed normally. The evidence does not determine whether the weak motion was ordinary vehicle/traffic state, the low effective cadence, or another cause. No harness repairs, further originals or controlled stepping followed.

All 56 original commands have complete ACKs, including quit. Three requests expired before dispatch with explicit effects=0; fresh captures and new IDs followed. No partial/UNKNOWN input or replay occurred. A read-only fetch for the nonexistent drive2 capture after a stale refusal failed without guest effect. An early generated-startup fetch also preceded readiness and failed; later read-only retrieval used a separate label. Both helper errors are retained. The cleanup template's generic UNKNOWN; never replay is not an input outcome.

Ordinary wall-clock rate and counter meanings

The measured interval was 03:13:07.539–03:13:37.543 UTC, elapsed 30,003.390 ms, during ordinary walking on the stairs. Native key holds and state/screenshot inspections occurred during the interval; their overhead is included. There was no Frozen toggle, controlled stepping, guest-clock scaling or scheduler change. The page remained visible to the headless browser.

The observer wraps the actual host _presentNow method, preserving its receiver, arguments and return value, and timestamps successful returns. The production method returns success when DirectDraw pixels upload or scheduled composition completes. It can publish repeated content and can combine work; therefore these events are screen upload/composition completions, not a validated count of unique game frames, compositor presentations or physical scanouts. The hook is restored afterward. Local preservation controls and a generated live Worker sample passed before the original.

Measurement Result
Successful screen upload/composition completions 77
Completion rate over measured wall time 2.566/s
Consecutive completion intervals 76
Minimum / mean interval 6.530 / 378.194 ms
Median / p90 interval 454.805 / 590.570 ms
p95 / p99 / maximum interval 669.700 / 763.900 / 763.900 ms
Naturally returned renderer submission delta 199 (2715 → 2914)
Transport flip delta 52 (711 → 763)
Owning-endpoint draw-call delta 11,487 (84,284 → 95,771)
Unique presented game frames / physical scanout rate unknown
Guest logical-update rate unknown

Percentiles use nearest rank over the 76 recorded intervals. Boundary gaps are not included in those intervals; the rate uses all 77 events and the full elapsed window. Renderer before/after snapshots are sequential naturally returned counters and can lag endpoint work. Submissions, flips and draws are different events, none converted into FPS. GTA2 has no registered validated logical-update marker in this staged app entry. Raw timestamps, both counter snapshots and analysis remain in rate-sample.json and rate-analysis.json. This one observed interval is not a clean benchmark or proof of general rate.

Actual WebGL ownership and tested identities

Original owning render Worker target: 0D112AD8E72FC49C581AEBB4B6AF86B1, session BC02953E49FAD594455A2E747959CD02, endpoint 1, legacy/webgl. At reverse1, its adapter is present, failed=null, ready=1, control error=0, completed sequence=13,004, batches=7589 and commands=250,769. Naturally returned endpoint stats report errors=0, fallbacks=0 and both direct software and fallback-software completions=0. Those are reported counters, not separate instrumentation of every GL call or parity acceptance.

Actual owning-context renderer string:

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

GL version: WebGL 2.0 (OpenGL ES 3.0 Chromium). Chrome was 151.0.7922.108, V8 15.1.206.16, headless, viewport 1024×768. This identity is from the owning endpoint, not the toolbar or the prior TR3 SwiftShader boat.

Before the original, actual production direct-initializer, atomic transport, executor/recovery, literal action, renderer key-state and duration controls passed. Up, Left, Space, Enter and Escape ran balanced trusted CDP pairs on about:blank in each browser; invalid holds produced zero events. The copied TR3 native preflight did not include Down or Right. Down passed the literal executor and production renderer/DirectInput state controls, but its later reverse action lacks the requested prior trusted-browser qualification. This closeout-discovered coverage gap further limits that action; it was not repaired by relaunching the original. Holds remain 50–2000 ms and the unmodified scene guard remains 12 seconds.

The generated Boids guest ran in a real Worker with an actual healthy WebGL owner, target B0777E81EC7047223A9477F2BF2E8C9B, reported errors/fallbacks 0 and the same Intel renderer. Natural draw calls advanced 536→603. Its colored boids and boundaries were personally viewed. A separate generated PE loaded original native COMCTL32, returned DllMain EAX=1 and showed the ordinary tabbed property sheet. A balanced click on its reviewed OK dismissed it. Both native captures are 1024×768 but differ in size/hash; both were re-viewed at original resolution after comparing identities, confirming presence then dismissal. Qualification browsers were fully cleaned before the original.

Staged source HEAD: 1d8d03ccabb91fde9b93c3625f6ecae62af56585. Relevant dirty/new-source patches and owned helper patches are retained; Git is the source archive. All three actual original Workers yielded Wasm SHA-256 07f1cc970038f8ab31313b8b200edf26caa63473f293f5fc937f5cfe5c5c7f6e, 2,181,179 bytes. This existing module was reused, without claiming staging HEAD as fresh module-build provenance. All 37 executing JavaScript source records from the three Workers matched exact staged bytes.

All 696 runtime/helper pins match remotely before the original and after exit. All 71 retained GTA2 media inputs, 26,207,806 bytes, match before/after. Installed gta2.exe SHA-256 remains 97ad743b6ec9ea1be95282053c3127084acd764a7f37a3e57e2268af52d02ae3; Miles remains 0974b244354a5d13e0711db15430c05f7949dc279b63897146f304d1401153fc. All 617 successful HTTP responses match exact files, including 251 original responses; there are no unmapped successful responses. Non-200 optional bundle/auth/build-info/favicon responses remain recorded. Individual fonts loaded. SHELL32/OLE32 used the recorded built-in stubs. Upstream provenance and historical native reference evidence remain unavailable.

Completed cleanup and remaining work

All 579 inventoried remote artifacts were retrieved and hash/size verified; duplicate modules have explicit retained-file mappings. Original Chrome exited normally at 03:21:47 UTC, code 0. An immediate inventory caught its controller finishing exit; the independent 03:22:01 check confirms all six tracked PIDs absent, six ports refusing, three profiles absent and all writer locks absent. The local messageboard watcher was stopped. No local guest runtime/source/build lane was used; initial offline helper adaptation errors were corrected before allocation and are not original-emulator failures.

Fresh boat bx_rs2buhtv used a persisted, fsynced UUID Idempotency-Key. Explicitly delegated normal stop stop_8f9f9ca61a57d939 completed at 03:22:44.561 UTC, forced=false, error=null. State is archived, active expiry null; initial expiry was 04:07:07.025 UTC. Allocation through retrieval and completed stop took about 15 minutes 38 seconds, within 35 minutes. Minimum observed local free disk was 2,091,184,128 bytes, above 2 GB. No old403 resume, held-risk retry, nested worker, source/media archive, automated risk flag or public deployment occurred.

Findings only are published; no production fix or full build is claimed. Meaningful sustained driving and a visually resolved steering change remain outstanding. The low observed upload cadence needs investigation before any ordinary-speed quality recommendation. No combat, audio, parity or full27 claim follows. Publication and ownership release precede result.json, written last, then this worker stops.