GLTron software fullscreen readiness: qualification stopped before original

No causal repair or successful original is established by this run. The prior original remains permanently closed. Its exact failed output guard operand is still unknown; fixture draw counters do not belong to that original.

Evidence: scratch/runs/20261118-gltron-software-fullscreen-display-readiness/. The immutable runtime remains f6962ba5b1cf2829a66f5cc4d80b293c30719b37 with module SHA-256 a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8. All 92 original media inputs were checked without modification. The original was not launched in this task.

Two authored software scopes used the qualified message-driven native GL PE 7457bf52fe3a09c5a09afb28d1e1fc4f5cf176f6eab00a7c369ec3b2c4eb027d and actual gl-gps=software. The first closed at the predeclared 1,520,000,000-byte disk reserve before completing a browser observation. The second retained positive production software observations: seven native draws, 3,718,176 native bytes, seven presentations, null GL renderer, and the owning software endpoint. A personally reviewed complete private display showed the native yellow triangle on blue. Trusted host fullscreen consent had an independent matching release with physical state zero; the first post-fullscreen current-display readback succeeded under the old validator. This did not reproduce GLTron's failure. The second scope then closed at the same disk reserve before guest Enter. No guest DOWN, Enter consequence, ordinary Quit or native guest exit is claimed. The later attempted local command was refused before writing a command file; there was no endpoint request after closure and no replay.

The candidate helper tools/gltron-display-readiness.cjs and its focused test remain uncommitted WIP, with exact new-file patches retained in evidence. No production renderer or shared admission file was edited here. The passive helper records the output guard's identity/dimension operands, window geometry, viewport and DOM visibility operands within the validator observation, preserves its original result/error, and emits diagnostics before failure returns. A startup-only candidate reuses the existing binding's ordinary-damage readiness check before the unchanged production validator; failure carries the same-call operands to the controller, which closes without a follow-up query. No crop, timestamp rewrite, unconditional admission, forced repaint, guest state change, clock, focus or polling change is proposed.

The native renderer test reproduces an ordinary queued resize with stale publication dimensions: the old guard refuses, startup readiness stays pending, and only a matching ordinary publication admits. Idle mismatches, hidden output, identity replacement and readback errors still refuse. All 43 focused tests passed, and the locked build passed. The candidate's actual Chrome old/new comparison and affected original remain unqualified; these checks are not substituted by the successful old-validator fixture fullscreen observation. No ready-to-publish code repair is claimed.

Profiles, cache and temporary output used exact published tmpfs helper 2a1afa6c4; raw scope data also used tmpfs, with raw XWD held in memory. Component updating and background networking were disabled. Eighteen actual runtime waits retained both EOFs; 69 recorded runtime PIDs were absent, both supervisors reported no leaks, profiles were removed after verified child waits, and both private displays closed. Selected named PNGs and unique logs were retained with exact gzip round-trip verification before removing owned tmpfs. The evidence includes Git pins, relevant binary diffs, new-file patches and recovery recipes rather than source-tree snapshots or media copies.

The original policy was fixed before any possible launch: one original, 180-second startup, 1,200-second total, 180-second Quit reserve and 20-second cleanup. Authored qualification has one fixed aggregate 600-second deadline; no scope/deadline was extended or reopened. Disk observation continues through sealing; final sampled minimum and monitor/board actual waits are in result.json. No RAM gate, workers, agents, boxes, Python, Wine or deployment were used. Box ID, expiry and stop receipt are null; state is NOT_CREATED.