Current-main original GLTron browser rounds, 2026-11-07

Original GLTron reaches its menu, ordinary START GAME, a living cycle and native steering on both requested software GL and WebGL, run serially in one fresh box. Both routes acknowledge a live S turn: direction 3 → 0 and trail offset 0 → 1, with winner -1, pause 1 and speed 8.5 before and after. Personally reviewed after images show the yellow trail bending beside the cycle. This establishes bounded browser round gameplay after ac1bcecee; it does not establish renderer parity, physical GPU performance, FPS, audible quality or full 27-title coverage.

Evidence: scratch/runs/20261107-gltron-current-browser-round/ (ignored). No production source was changed. This was one bounded worker task, without nested workers, guest-state/configuration edits, forced rounds, clock changes, private game handlers, frozen stepping or counter arming.

Runtime and actual ownership

The tested source is main f676b9335b432a8e0e21b35d6ce836360819c2c8, which contains ac1bcecee. relevant-source.patch is empty; the existing untracked lib/wininit-boot.js is preserved separately as a new-file patch. The actual browser compilation and instantiation both report module SHA-256 f9650607181301b5d2a9990420f437bd9f1ca10d40ac73067339f07499d53121, 2,182,500 bytes. The identical retrieved module is retained. lib/host-audio.js is SHA-256 e2ae7f782e7fbeb246471acfb3a93d7ae80eec9b195e44f02072310e81373d15; host.js is f8611a65d9f216c1c8c3c2a185715badd9442b273d770b156fef55e42a565766.

Both ordinary registered APPS.gltron launches use all 92 original files (4,968,893 bytes), with the original EXE SHA-256 85fb5184262490f0b5967c2b12ee27c77da058364e594b0fc86be7b5814e40ec. Media paths, sizes and hashes are in media-refs.json; media was not copied into evidence. Every one of 234 served file reads per route matches its pins, including all 92 originals. Production/media pins are unchanged after both runs. The software seal has 720 files; the WebGL seal adds three run helpers. One external WebGL reader helper was revised to include device ownership fields after page launch; its exact old/new hashes are explicitly recorded in retrieved/postflight.json. No production/media file changed with it.

Chrome is 151.0.7922.108, Node v24.22.1, Puppeteer 24.37.2. Both pages have WINE_THREADS=false, no guest Worker, cooperative thread backend, actual CPU thread 1, visible/focused screen, frozen=false and zero host steps. Normal production elapsed-time scheduling and audio were used. The diagnostic frozen-status tick field does not mean frozen mode was enabled.

Requested route Actual owning renderer/context Round evidence
Software GL Main GLRenderHost.Bridge.local / OpenGLHostBridge; owning exports match actual CPU exports; context 1 has software=true, gl_sw_enabled=1; no WebGL context Reviewed main/GAME menus, paused arena, living cycle/trail, native live S turn, crash/winner and ordinary Enter restart
WebGL Main OpenGLHostBridge._webglBackend is WebGLBackend; its .gl is the observed WebGL context; current context 1, owner/context map [[0,1]]; owning exports match CPU, gl_sw_enabled=0 Reviewed main/GAME menus and confirmed paused arena, native unpause and live S turn, changed yellow trail, original PLAYER 3 WINS result

The WebGL context reports WebGL 1.0 / OpenGL ES 2.0 Chromium and ANGLE (Intel, Mesa Intel(R) UHD Graphics 620 (KBL GT2), OpenGL 4.6 (Core Profile) Mesa 23.2.1). This is browser-reported context identity. The toolbar's WebGL label was also visible during the software run and is not used for GL backend attribution. The additional empty, nonmatching bridge belongs to the cooperative sibling; it is not substituted for the matching main render owner.

Reviewed gameplay and ordinary controls

Software captures retrieved/software/startup.png, game-menu.png and arena-paused.png establish the main menu, GAME/START GAME and the original paused arena. Each dependent Enter followed personal review of the preceding scene. living-before-turn.png shows the moving yellow cycle and its new straight trail, without the pause message.

The first human-paced A/S attempts were acknowledged but their subsequent captures show crashes. They are not claimed as surviving steering. Ordinary Enter restarts are retained, including the original WINNER: 4 result. A final prompt native Enter/S sequence then samples only the existing original player layout through guest_to_wasm, reusing readLiveState from tools/gltron-logical-fps.cjs without importing or arming its counter code. The original executable/layout controls are reused unchanged. native-steering-result.json proves the live direction/trail transition. steer-live-after.png visibly shows its yellow corner. Old crash/winner text persists across software restarts, so text overlays alone are not used to identify player liveness. The external Puppeteer connection changed the software viewport from 1200×900 to its 800×600 default; this is recorded in the captures and is not a rendering comparison.

WebGL webgl-startup.png and webgl-game-menu.png establish its separate ordinary launch. The first capture named webgl-arena-paused still showed old menu pixels; no dependent input used it. The later personally reviewed webgl-arena-confirm.png shows GAME IS PAUSED and the yellow cycle. A native Enter/S sequence unpauses and turns only after the read-only live-player guard passes. The before capture still retains pause pixels despite its live guest sample; the after image visibly shows an unpaused cycle and bent yellow trail. retrieved/webgl/native-steering-result.json supplies the matching live transition and actual WebGL device/context identity. webgl-round-result.png was personally reviewed and shows PLAYER 3 WINS. This connection preserves the existing 1200×900 viewport.

All 10 software and four WebGL native key pairs have trusted screen-targeted DOWN/UP events, with no held keys at termination. Both helpers release keys in finally; the route owners also perform normal physical release during closure. Every actual command ACK succeeds. Neither original route has an UNKNOWN command or runtime first-fault record; no uncertain input was replayed. No five-second measurement handoff or rate interval was used.

Setup boundaries and verification

Two pre-Chrome pin checks caught npm replacing a staged CLI dependency and the package lock. Their failures are preserved. Browser dependencies were isolated outside the staged runtime and all staged pins restored before Chrome. The first Chrome attempt then exposed a helper mistake: GLTron was absent from the local selector, selection silently fell back to Notepad, and that unrelated load failed. Its selector, logs and zero-exit normal closure are preserved under pregltron-picker-failure; no GLTron executed there. The helper now exposes the existing registered GLTron entry as a local DOM option and asserts its selection before clicking the ordinary Launch button. No app registration or guest configuration was changed.

A subsequent corrected-pin launch wrapper failed while writing its metadata after spawning the driver (selector was undefined). That exact exception is retained; the driver was observed running and was not launched again. The resulting single software GLTron launch and later serial WebGL launch are identified by their own browser identity records. These known setup failures are distinct from input outcomes.

Local helper syntax and structural original-registration checks pass. unchanged-controls-verification.json confirms 301 source/module pins identical to the preceding passed clock/audio/callback investigation; those controls and fixture hashes were reused rather than repeating the CLI census. The existing module is unchanged and no code commit/build is required for this documentation-only result. Git SHA, relevant diff, untracked source patch, owned helper patches and dependency/media/module pins provide recovery without a source-tree or media archive. GLTron has no candidate-manifest entry, so this remains an investigation note rather than an unrelated dashboard card. Own-scope evidence and whitespace checks pass. The global evidence checker reports seven existing containment errors in other runs, none in this run.

Retrieval and shutdown

All 150 artifact references / 143 unique contents / 12,750,040 referenced bytes were retrieved and size/SHA-256 verified before requesting normal box shutdown. Both guest browsers exited zero with null signal: software Chrome 20791, WebGL Chrome 23907. Both PIDs are absent, their HTTP ports 40135/34299 refuse connections, no owned browser-driver/profile process remains, and the owned profile was removed after those checks.

The detached independent 25-minute watchdog and normal-stop delegate were armed and observed before allocation, with durable heartbeats and immutable deadline records. Both delegates exit zero, the supervisor has no remaining children, and maximum watchdog heartbeat gap is 0.782 seconds. The local messageboard watcher is stopped. Launch disk exceeded 5 GB locally and remotely; there was no RAM gate. Normal stop followed verified retrieval promptly, without idle handoff. Box bx_xqjv9b99 is archived, current expiry null; original expiry was 2026-11-08T00:04:54.488Z. Stop stop_65b0a62f9f1a65d6 completed 2026-11-07T23:50:22.632Z, forced=false, error=null. No resume, delete, source archive or public deployment occurred. This task ends with the two bounded gameplay cells; any longer survival, visual-defect repair or broader qualification is separate future work.