Current TR3 Jungle movement after restored headroom — 2026-11-08

Current Software D3DIM shows persistent bounded Jungle translation and a separate heading change. Current WebGL shows a persistent heading change, but terrain-relative translation remains unproved. No emulator regression was established and no emulator code changed. Full27, backend parity, FPS, audio and level completion remain open.

Evidence is in ignored scratch/runs/20261108-tr3-current-movement-after-headroom/. This follows tr3-current-native-gl-gameplay-20261108.md and keeps its accepted WebGL world/pistol draw separate. The older November 7 software result is not substituted for this run.

Route and personally reviewed pixels

One fresh UUID-idempotent box ran WebGL, drained completely, then ran software. Each backend used one original executable launch and unchanged native welcome OK → setup OK → Game → visibly open New Game passport → Jungle. Setup retained Primary Display Driver, Direct3D HAL, Hardware Acceleration, 640×480 High Colour 16 Bit and ZBuffer. Ordinary host renderer queries selected the backend. No media/config/save/clock/seed edits, forced scene, private handler, guest memory writes or repeated launch were used.

Raw captures were opened at their original 945×889 resolution. All 38 retained PNGs pass chunk CRC and complete PNGjs pixel decoding. The game buffer is 640×480. Review sidecars bind the exact image hash and current owner to each input. Captures named startup were superseded by personally reviewed fresh welcome captures; capture existence alone is not acceptance.

For WebGL, jungle-before.png still includes startup animation and the Controls overlay. Left2000 followed by four seconds idle gives left-after.png; six more seconds idle gives left-persistent.png, a stationary side view beside the moss wall and tree. The initial change is not isolated from startup and is not accepted as movement or heading by itself.

From that settled image, a separate Right2000 produces right-after.png and, after eight more seconds idle, right-persistent.png. Lara's torso/backpack and feet retain a changed orientation while the same moss wall, tree and starting rock remain fixed. This supports a bounded heading change independent of forward travel. No angle is measured.

Two separately inspected Up2000 pulses produce forward-after.png and up-after.png, each followed by ten seconds without input in forward-persistent.png and up-persistent.png. The viewpoint changes substantially, including camera follow around Lara. These images do not isolate an unambiguous persistent terrain-relative displacement, so current WebGL translation is not accepted. The last Up hold's passive samples all retain slice sequence 8953; the earlier Up samples span 8419–8425. These are sparse host observations, not guest keyboard-poll, simulation-step or FPS measurements. They explain why physical ACKs cannot establish movement. The exact remaining blocker is insufficient visible displacement evidence within the qualified 50–2000 ms physical-input envelope, with camera follow confounding the views. No longer hold, timing patch or blind repetition was used to bypass that limit.

For software, jungle.png shows the settled side view and Controls overlay. Space300 and four seconds idle give armed.png; another five seconds idle gives armed-persistent.png, with raised pistols and unchanged terrain. Up100 gives forward-after.png, retained six seconds later in forward-persistent.png. That first change includes camera follow and is not the sole translation proof.

From this already settled downhill view, a second separately reviewed Up100 gives step-after.png. Lara stands farther along the same foreground rock, with the large boulder to her left and the downhill stump/corridor still identifiable. step-persistent.png, six seconds later without input, retains that changed location and stationary armed stance. This is the bounded visual translation result; no world-space distance is claimed.

A separate Left300 then gives heading-after.png: Lara faces toward the large boulder instead of down the corridor. heading-persistent.png retains this orientation after six more seconds idle. The camera follows the turn, but the changed relationship between Lara's facing direction and the boulder/corridor persists without forward input. All reviewed gameplay remains in the same living Jungle scene: no death pose, title return, New Game repetition or respawn was observed or used as displacement evidence.

One stationary software Control1000 was attempted after that armed scene. fire-after-held.png and fire-after.png show armed poses, but no unmistakable muzzle flash, impact, ammunition change or target consequence. Firing and its consequence remain unverified. No target was visible and no second shot test was attempted. WebGL firing was not attempted.

Actual owners and current module

Published staging Git commit is 0a38cf48438096482d059c5d8f27c048474c83b9. Individual tracked runtime files came from that commit's Git blobs; peer WIP was excluded and its relevant binary diff retained. Existing non-Git helpers and module pins are recorded separately. There was no rebuild and no source-to-module equivalence claim.

Actual guest and render Worker bytecode is 2,182,452 bytes, SHA-256 7ec49cbbae4cb177909fd8fd9d3599713ee5f2a5ffb7f7e8c890848a540052bf. All 48 WebGL and 42 software module/checkpoint references match that pin. All 189 WebGL and 181 software complete successful HTTP responses, plus four loaded host scripts per backend, match staged pins. Both final drains verified all 603 runtime pins unchanged. Ten TR3 fixture files remain referenced by existing path and SHA-256; no media is repeated in the evidence.

WebGL guest target is 504B663F0B93C039A4E77A75EB532A93, endpoint 1, slot 0, GPU enabled. Actual render owner E4B813D322B9138F247EB27E5DC0EB57 owns legacy/webgl, an observed OffscreenCanvas WebGL2 context at 640×480, and advancing production commands. The observed renderer string is ANGLE (Intel, Mesa Intel(R) UHD Graphics 620 (KBL GT2), OpenGL 4.6 (Core Profile) Mesa 23.2.1). Physical GPU remains unknown. Reviewed WebGL checkpoints have errors=0, fallbacks=0, software completions=0 and GPU-fallback software completions=0.

Software guest target is 512289E41E22286BBCEE270B0900B2C9, GPU disabled. Actual owner EC28526E011065C0EC04D3267BED7E1D owns endpoint 1, legacy/software, with no GPU adapter or observed GL context. Its direct native software completions advance from 137,838 at persistent weapons to 176,592 at persistent translation and 198,912 at persistent heading. Shared control remains ready=1/error=0, failed=null; guest fallback and GPU-fallback software completions remain zero. Sequential samples are not atomic pairs. Chrome is 151.0.7922.108, with production guest Workers and native viewport.

Guards, retrieval and normal shutdown

Reused local controls passed native input before graphics, one-way actual owner binding, replacement-owner rejection, stale zero-effect refusal, independent release despite slow sampling, atomic numbered publication, exact ACK identity, duplicate rejection and sticky UNKNOWN closure. Each backend also passed actual trusted blank-page native key pairs and invalid-duration zero-effect tests.

The original queues contain 37 exact complete ACKs: 20 WebGL and 17 software. Three are stale refusals before dispatch, effects=0. Continuation used fresh captures and new IDs, never consumed-ID replay. No original UNKNOWN or partial action occurred. A proposed Shift+Up combination was rejected by local helper validation before locks/publication/input because the released helper does not support that combination. No helper relaxation or runtime change followed. The read-only local retrieval helper was changed to fetch the three image sidecars concurrently, reducing review latency while preserving all guards.

The first immediate WebGL drain saw its controller finishing after Chrome exit; a later read-only drain and both final drains prove all owned controllers, Chrome/profile processes, ports, profiles and writer locks absent. Chrome exited zero at 05:53:44.184 UTC (WebGL) and 05:57:36.898 UTC (software).

All 415 artifact references, 369 unique hashes, were retrieved and independently size/SHA-256 verified at 05:58:04.181 UTC, before stop. Unique modules, captures, logs and exact command/ACK records are retained. The independent observable 25-minute watchdog was armed before allocation. Box bx_yh7snky3 is archived, final expiry null; initial expiry was 06:12:58.222 UTC, independent deadline 06:12:58.011 UTC. Delegated normal stop stop_fca890eb187e1b98 completed 05:58:05.036 UTC, forced=false, error=null. Both delegates exited zero and the supervisor records no remaining children. Recorded disk samples remained above 5 GB; final sample was 6,290,513,920 bytes. No RAM gate, HTTP403 resume, box deletion, source/media archive, deployment or nested worker occurred.

Only this findings note is committed. Next work is an ordinary-input method that can isolate current WebGL terrain displacement within the qualified input limits; no emulator repair is justified by this run alone.