GLTron software route: fixture passed, original display admission closed

The single unchanged GLTron original did not establish normal pause, either manual turn, or ordinary Quit. It reached the textured root menu and received one ordinary host fullscreen-consent click, independently released to physical state 0. Before any guest key, its first game-frame preparation permanently closed on Current display output changed or is cropped. No endpoint was requested after closure, and no original was replayed.

Evidence is scratch/runs/20261118-gltron-software-paused-turns-native-quit/, with result.json written last. The earlier WebGL pause/two-corner result and the separate earlier WebGL Quit result remain separate runs.

The immutable runtime graph is f6962ba5b1cf2829a66f5cc4d80b293c30719b37, with recovered and served module SHA-256 a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8 (2,418,710 bytes). All 92 registered media inputs matched before and after. The actual selector was gl-gps=software; component updating and background networking were disabled. Optional child Debugger inspection was omitted, so independently executing child module bytes and roles remain unknown.

One focused native GL software fixture used the unchanged message-driven PE SHA-256 7457bf52fe3a09c5a09afb28d1e1fc4f5cf176f6eab00a7c369ec3b2c4eb027d. Personally reviewed whole-display PNGs show its yellow triangle on blue, then red after ordinary Enter. Independent matching releases read physical state 0; ordinary Escape produced actual guest exit 0. Current-display readbacks and native transfer/publication receipts were retained. The owning gl-gps endpoint requested software, and its production executor receipts contained 13 native WAT draws, 3,718,176 native bytes and 13 presentations, with null glRenderer. This proves software work in the fixture, without inferring it from a UI/backend label or claiming backend parity.

The original's retained startup counters establish software allocation, clears and presentation, but the sampled native draw count was still zero. Its later textured menu was personally reviewed in original/wholehost/consent-001.png. No later positive native draw snapshot was retained before closure; the fixture's positive counters are not assigned to the original.

The original closed at 2026-11-18T11:46:56.395Z. After host consent, five passive display-readiness replies reported an ordinary display transaction in progress at publication 41. The next request reached the production renderer's _validateCurrentDisplayCapture and threw its output identity/dimension/crop guard. This happened inside prepare, before release prearming or requesting the game-frame ticket. The exact failing subpredicate was not retained: fullscreen transition timing is a candidate explanation, not an isolated cause. The fail-closed validator was preserved. No speculative production repair or successful affected-original validation is claimed. Focused native current-display and GL servicing tests passed; the existing qualified Space/Enter/A/S sequence and bindings were retained but never sent here.

Disposable browser profiles, cache, temporary files and all raw scope output used a unique directory under /dev/shm. After actual browser/native/holder waits and EOFs, selected wholeframes and unique logs were copied to disk; compressed logs were verified by exact decompression. Unselected raw pixels are explicitly unrecoverable. Git plus relevant binary diffs supplies source recovery; no source-tree snapshots or media copies were retained. Published tmpfs helper 2a1afa6c4 became available after launch; the already frozen run-local operator was preserved.

Bounds were one 300-second fixture, one 1,200-second original with a 180-second Quit reserve, 180-second startup and 20-second cleanup. Neither scope was extended. The original host browser exited 0; the failed driver and holder exited 1. Both supervisors reported no leaks, recorded runtime PIDs were checked absent, and both private displays closed. The finite independent disk guard continued through drains; its final sampled minimum and actual wait are recorded in the sealed result. No RAM gate, game/config/save/state/ clock/focus/polling edits, solver, OCR, autoplay, agents, workers, boxes, Python, Wine or deployment were used. Held/foreign work was untouched. Box ID, expiry and stop receipt are null; state is NOT_CREATED.