Tomb Raider III software Jungle movement and weapon draw — 2026-11-07
Personally reviewed settled Jungle translation and a separate weapon draw in
original TR3 on the supported production software renderer. This advances
GLD3D-CORPUS-27-20261006 beyond the WebGL-only d5c39cdf2 result. No emulator
defect was demonstrated; this publication changes findings only.
Evidence: scratch/runs/20261107-tr3-software-gameplay/. One fresh boat, one
original executable launch, ordinary native browser input, no local runtime,
source or build lane, no nested worker, deployment or source/media archive.
Original media, config and saves were not directly edited or substituted.
Personally reviewed route and result
The demo welcome OK and setup OK were selected from fresh captures. Setup showed Direct3D HAL, Hardware Acceleration, 640×480 High Colour (16 Bit), and ZBuffer; the host GPU selector showed Software. The guest's Hardware Acceleration option selects its Direct3D API, while the supported host query selects software replay. The red Game passport was opened with balanced Enter300; a fresh capture visibly said New Game before a second Enter300 selected it. Jungle's Controls overlay documented Up Run and Space Arm.
The first Up2000 moved Lara to the spike-lined lower clearing. The next idle capture returned to the title. The cause of that return was not instrumented; this segment is not accepted as settled movement and does not establish a software defect. From the personally reviewed title, ordinary Game → New Game was selected again inside the same running original process. No EXE relaunch, unknown-command replay, memory reset or save/config edit occurred.
The successful bounded sequence is retained separately:
review/jungle-before-arm.png: empty hands and holstered pistols at the starting ledge, viewed side-on with the Controls overlay.- Balanced Space300, then 4000 ms without input:
review/jungle-armed.pngshows both pistols raised and the overlay removed. Another 4000 ms without input producesjungle-armed-persistent.png, retaining the stationary weapon-ready pose. This is weapon draw, without firing. - From that freshly reviewed armed scene, balanced Up100, then 3000 ms without
input:
review/jungle-short-forward.pngshows Lara farther down the starting slope with the camera behind her looking downhill. The right wall, foreground rock and downhill corridor differ from the starting side view. - Another 6000 ms without input produces
review/jungle-forward-persistent.png: the changed terrain/camera position persists with Lara stationary and pistols raised. This establishes translation beyond an isolated running animation.
Eighteen numbered commands have complete ACKs. Two expired requests (welcome OK and the first proposed Up) were refused before dispatch with zero effects. Fresh captures and new IDs continued on the same queue; consumed IDs were never replayed. No partial or UNKNOWN action occurred. Each ordinary input followed personal inspection and passed the current-app and 12-second scene guards.
Actual software owner and execution identities
The ordinary URL is ?d3d-renderer=software. A direct live page probe confirms
WINE_D3DIM_GPU=false, WineD3D.renderer=software, and selector software.
The actual process render factory owner is Worker target
169A9316F2274904A1B1C547FE3C9220, session
FFDEEC2F9B9FDF790CB4829F7C833696, endpoint 1, legacy/software.
Production replay calls ex.d3dim_worker_draw and only then increments
evidence.d3dimSoftwareDrawCallsCompleted.
| Reviewed checkpoint | Direct owner software completions | Completed sequence | Guest-reported software completions |
|---|---|---|---|
| Before Space | 307,416 | 18,054 | 307,158 |
| After Space | 386,010 | 20,568 | 385,969 |
| Persistent weapons / before short Up | 401,103 | 22,808 | 401,057 |
| After short Up | 419,721 | 25,489 | 419,702 |
| Persistent translated position | 444,610 | 28,822 | 444,558 |
Samples are sequential, not atomic pairs. At every reviewed gameplay checkpoint, the owner has no GPU adapter, failed=null, ready=1 and error=0. Guest endpoint statistics report GL triangles=0, GL renderer/version=null, fallbacks=0 and GPU-fallback software completions=0. The actual software owner, absence of an adapter and advancing native completions establish software execution with no WebGL draws on this route. No separate GL-call instrumentation was installed.
Source staging commit is 0e37d2fd73ee8a03c843e803d461e6a81f723ff5, with relevant
uncommitted and new-source binary patches retained. All 581 provisioned runtime
file pins match remotely before and after execution. Actual guest and render
Worker bytecode SHA-256 is
762e723515df6ce8c197fd0d23e0b84869657b0b49e673ad4ac2c48f78b8991f
(2,181,062 bytes), matching the provisioned existing standard module. Source
identity and reused module identity are recorded separately; no rebuild or
source-to-module equivalence claim is made.
A direct CDP debugger probe records actual parsed script sources, identities and hashes: all 111 external executing scripts match pinned bytes; four inline scripts have separate hashes. All 169 successful original HTTP responses match pinned bytes; five non-200 responses are retained. Chrome is 151.0.7922.108, isolated guest Workers, 1024×768 viewport and 640×480 game buffer.
Preflight, provenance and closeout
Before the original, nine transport, seven actual-executor/recovery, balanced production-renderer keyboard and invalid-duration controls passed. Holds 0, 49, 2001, 3500, -1, NaN and 100.5 fail before locks, command publication, admission, state reads or key effects. Allowed holds are 50–2000 ms.
A generated PE with separate ILT/IAT and ordinary property-sheet flags loads the
exact retained native COMCTL32 through its dependency URL. The served DLL hash
is 28a41e6a35c4509924a32a99841a815cf49e1d3ee6912ec26a823e45472edfa4;
DllMain returns EAX=1. Its explanatory tabbed sheet, OK, Cancel and disabled Apply
were personally reviewed before normal close and clean process/port/profile/lock
checks. TR3 itself resolves its eight imports through built-ins and does not use
native COMCTL32 on this route.
All ten TR3 media pins exactly match the retained current-browser fixture record.
The 920064-byte executable SHA-256 is
8c7b25460f62c180ae675c2696ac66c8f97948a16458d0598ba72794ddcc7814.
provenance.json, complete upstream attribution and historical native evidence
remain absent. Exact retained local identity does not recover upstream provenance.
No download, repeated media in evidence, replacement fixture or guest-byte edit
occurred.
All 193 terminal inventory files were retrieved or matched locally with exact
size/SHA-256. Original Chrome exited zero at 01:52:25.447 UTC. The immediate
inventory still saw the controller; the later independent check at 01:53:00.781
found both generated/original controllers and Chrome absent, all four ports
refusing, both profiles absent and no profile processes or writer locks.
An initial cleanup assertion accidentally matched its own Python command text;
the retained diagnostic fixes the read-only search to compare individual argv
entries. This was a run-helper assertion, not an emulator defect or risk flag.
The generic cleanup pending:UNKNOWN text is a template, not an action outcome;
the eighteen complete ACKs establish the actual command outcomes.
The UUID Idempotency-Key was persisted and fsynced before creation of
bx_9h56d6hu at 01:46:50. Initial expiry was 02:46:50.021 UTC.
Explicitly delegated normal stop stop_fd8c5b9c1c3a3483 completed at
01:53:05.408 UTC, forced=false, error=null. State is archived, active expiry null.
Allocation through completed stop took about 375 seconds, within 35 minutes.
Disk remained above 2 GB. No automated risk flag, held-risk retry, HTTP403,
resume or original relaunch occurred.
Artifacts and publication records precede result.json, written last. No code
repair or build was needed. Combat, audio, FPS, parity and full27 remain untested.
The bounded software movement/action task requires no further execution.