Tomb Raider II bounded browser gameplay — 2026-11-07
Ordinary forward translation and a distinct weapon-draw action personally
reviewed in Venice, with owning WebGL Worker evidence. This closes the narrow
movement/action gap left by 9218af507 under GLD3D-CORPUS-27-20261006. No emulator
defect was demonstrated and no production code changed.
Evidence: scratch/runs/20261107-tr2-bounded-gameplay/. One original on one fresh
boat, bx_renx76cm; no second original, replay, old403 resume, nested worker,
media replacement, guest/save/config edit or shared local runtime/build use.
Before allocation/original, generated controls passed: nine atomic transport cases, seven actual executor/recovery cases, balanced production-renderer and DirectInput keys, and invalid durations 0, 49, 2001, 3500, -1, NaN and 100.5. Invalid holds reject before writer lock, command/ID publication, admission, app reads or key events; valid bounds are 50–2000 ms. The run-only adapter adds validation at both request entry and executor entry. The corrected generated property sheet has separate ILT/IAT, ordinary flags, and a personally reviewed tabbed page with explanatory text, OK, Cancel and disabled Apply. Its Worker log confirms native COMCTL32 loading and successful DllMain.
The original began at 00:55:40.073 UTC. The current Graphics page was personally reviewed: Hardware 3D Acceleration, full-screen 640x480 High Color and Z Buffer. Ordinary OK received a complete ACK and reached Venice. Each input was based on its current personally reviewed capture, within the adapter's scene guard:
review/venice-idle.png: settled initial alley, empty hands and Lara facing the orange lattice window uphill.- Balanced Up 1600 ms, then 900 ms:
review/forward-settled.pngshows running gait; this intermediate frame alone is not called settled proof. - A separate 5000 ms without input produced
review/walk-after-idle.png: Lara stands idle farther uphill with the nearby brick corner shifted left, changed floor position and closer slope/window. Facing is unchanged and no turning input was sent. The persistent spatial change after movement settles demonstrates translation. Up is running; Shift-walking was not tested. - Balanced Space 300 ms, then 5000 ms:
review/weapons-drawn.pngshows raised arms with visible pistols and a health HUD. After another 6000 ms without input,review/weapons-ready.pngretains the distinct weapon-ready pose. This demonstrates draw weapons, not firing or combat. Alt jump was not sent.
The original PC manual
identifies Up as forward running, Space as draw weapon and Alt as jump (pages
2, 5 and 7). controls-reference.json records the source. All six original
queue commands have complete ACKs; no stale refusal or partial/UNKNOWN input
occurred. No combat, damage, kills, audio, FPS, backend parity or full27 claim.
At weapons-ready, the returned owning endpoint counters identify legacy
WebGL endpoint 1: 17142 draws, 54103 draw calls, 72364 triangles, errors 0,
fallbacks 0, software draws 0 and fallback-software draws 0. Direct CDP inspection
of the actual D3D render Worker identifies endpoint 1, backend webgl,
failed=null, ready=1, error=0, completed=935, batches=633 and commands=54656.
render-evidence.json distinguishes these direct control/identity readings
from endpoint counters returned through the guest Worker; no separate GL-call
instrumentation was installed. Renderer: ANGLE Intel UHD Graphics 620,
WebGL 2.0; Chrome 151.0.7922.108. Both guest and render Workers supplied actual
Wasm bytes with SHA-256
6cd858302895de91df5af17e3a3742281e31f369ddb6528d7b96252b3423caac.
Staged source commit: cdf3d3f9aebf6c616f3ba7d3f00d30d7a642d78d, relevant dirty
and new-source patches retained. The existing module was reused without a
local rebuild; source staging identity and actual loaded module identity are
recorded separately. All 574 remote file pins match before and after, and all
171 successful original HTTP responses match; five non-200 responses are
retained in the audit. No source/media archive was created.
Exact supplied fixture hashes were verified against retained evidence:
Existing path under test/binaries/ |
Bytes | SHA-256 |
|---|---|---|
candidates/tomb-raider-2-demo/game/TOMB2.EXE |
909824 | 0811efa84bfab8c458b9e29976088152db4c14601a99ca4910c75a2d009233aa |
candidates/tomb-raider-2-demo/game/DATA/DEMOPC.DAT |
2064 | e1bdd5776e1946420e0eeb9e1a4da12f16ea2bbfb7c3909b414372b100ff51bd |
candidates/tomb-raider-2-demo/game/DATA/BOAT.TR2 |
5920318 | e710716fcaf039fb3462e26e33f0ead8eb45bd01c8bbfa036c455a1d19823d9e |
dlls/comctl32.dll |
577808 | 28a41e6a35c4509924a32a99841a815cf49e1d3ee6912ec26a823e45472edfa4 |
Upstream provenance.json and historical native evidence remain absent.
Historical attribution is tr2_demo_02.exe from Tomb Raider Chronicles; local
hash agreement does not recover upstream provenance.
All 95 final remote artifacts were retrieved or matched locally. An initial
retrieval stopped on a local exclusive-create collision with an already fetched
generated log; remaining retrieval verified existing bytes and downloaded only
missing files. It did not repeat guest input. The reused post-pin checker prints
an obsolete constant originalLaunches:0; the launch/queue records establish
exactly one original, and the checker establishes file hashes only.
Chrome exited normally with code 0 at 00:57:41.143 UTC. Runtime processes are
absent, both ports refuse connections, profile and writer lock are absent.
The task's explicit normal-stop requirement delegated this box's stop:
stop_9f9533f82fb8a8b1 completed at 00:58:20.709 UTC, forced=false, error=null;
box archived, active expiry null. Allocation through stop took about 250 seconds,
original start through stop about 161 seconds, below 35 minutes. Local free disk
stayed above 2 GB. No automated risk flag occurred. Findings only are published;
no production-code build claim. Remaining gameplay areas need separate evidence.