GLTron released atomic precheck, 2026-11-14

Code a918f0b61854121fddce5d9d5671b9f0f6d1fc08 is published on main. Actual Chrome controls and the sole fresh original demonstrate that a new held paint before the ticket is a passive precheck. The original observed five such refusals, then captured the complete native fullscreen title/menu and received one permanent binding. One finite Return opened Game. The next capture encountered a post-ticket hold and closed permanently. Two intact-cycle corners, a natural round outcome, human identity and native Quit remain unproved.

Evidence: scratch/runs/20261114-gltron-released-atomic-precheck/result.json. The predecessor report 7d00ada5 remains unchanged and describes its own unpublished preparation. This task adopted and reviewed its released three-file WIP, added boundary controls, completed production validation and published it.

Transaction and fixed bounds

The asynchronous identity path recovers the actual guest and renderer module bytes and validates their hashes, targets and object identities. It rechecks the actual Worker instance and owning renderer after module recovery. The final synchronous page transaction validates the same page owner/output, visibility, paint and current display receipt, checks the original startup deadline, then calls requestCurrentDisplayCapture. There is no await between those checks and the request. Existing armed/pending tickets refuse before passive classification.

Only a positive NOT_READY response with a fresh matching nonce and requestCalled:false, ticketCreated:false, inputDispatched:false can remain in the first capture's fixed startup window. Each new probe repeats identity validation. These are nonmutating observations, not guest input. A request exception, failed ticket, cancellation, ambiguous response, UNKNOWN or post-ticket hold closes the chain. An actual failed ticket now has the distinct terminal CAPTURE_TICKET_FAILED classification; it is never called an unsent precheck. Later captures have no startup-wait privilege.

The overall startup bound stays 90 seconds; CDP and capture bounds stay 3,000 ms. The existing 5-second precision and 30-second menu freshness, finite 1,500 ms maximum gestures, independent 1,750 ms UP, single-use nonces and mechanical 180-second Quit reserve remain. Identity validation cannot refresh image age. Original publication/display, request and readback timestamps remain attached to the admitted bytes. No visibility/hold guard, repaint/swap/hold release, raw slot, guest clock/configuration/state write or interpreter change was used. Host, index, page viewport and renderer files were read-only for this task.

Actual controls and publication

chrome-v3/report.json records actual production Chrome, WebGL Worker pixels, the production compositor and an attached live WASM module. The producer's normal begin message is introduced after initial readiness during asynchronous Debugger.getScriptSource validation, before page arm. Exact committed ee9c89fe closes with zero tickets and zero input. The candidate records 12 passive probes with direct unchanged request counters and no ticket, then one unheld immutable current-display ticket under the same owner. Its PNG pixels exactly match the actual Chrome compositor screenshot.

The same run passes sustained ordinary Worker paint with an actual fixed 1,000 ms test deadline (expiry observed 5 ms after that deadline), four cyclic Worker paint intervals, owner replacement during module validation and after the final awaited page read immediately before arm, cancellation of a real pending ticket, and a post-ticket late hold. Closed chains refuse reopening. Actual 3-second CDP UNKNOWN remains closed after its late reply and cannot send input. DOM overlay, modal, popover, hidden/clipped/replaced output and the ordinary trusted host fullscreen phase pass their respective controls.

Both failed development controls are retained with actual waits and both EOFs. chrome-v1 passed the causal late-hold comparison but its 250 ms timeout fixture expired during initial setup. The fixed control bound became 1,000 ms with a longer ordinary producer interval. chrome-v2 passed the subsequent negative controls, but its three-frame cyclic hold ended during module validation before an arm could observe it. The final cyclic workload uses twenty-frame intervals. No timer or sleep is treated as hold-release evidence; actual Worker end messages and current readiness establish natural release.

Final focused controller/loopback/actual-child tests pass. Five renderer test files pass all 20 entries. The locked full build passed in 29.116 seconds. Blank-page Chrome controls verify trusted finite keys and mouse, including 1,500 ms pulses and independent 1,750 ms EOF UP. Final Chrome pins match the three committed code files. Explicit own-path commit and strict pull/push completed before the original.

Sole original and terminal result

Pegged's actual native/watchdog/supervisor-child waits with both EOFs, heavyReleased and absent exact processes satisfied the amended prerequisite. No fourth ancillary wrapper wait was required. The historical StarCraft outer wait/EOF UNKNOWN remains unchanged. The new original had an observable independent prearmed 600+20-second watchdog, default 1,000-step slices, real clock and all 92 existing media hashes. Git SHA and relevant binary patches preserve source identity; there are no source trees, media or WASM copies.

original/host-bootstrap.png was personally reviewed before its single normal trusted fullscreen click. The exact pre-click visibility refusal was touch-cursor; this does not identify the older f693 run's unknown sole clause. DOWN, UP and click were trusted and targeted the consent button, with no canvas mouse event. The same wrapper became fullscreen, the strip disappeared and output visibility passed.

Initial readiness was publication 153. Five nonce-distinct pre-ticket held observations followed, all reporting zero request/ticket/input effects. Natural readiness then admitted original/startup.png, publication 158. It depicts the complete native GLTron title/menu with GAME selected above VIDEO/AUDIO/QUIT, without the strip or touch cursor. Display time was 22:31:28.820 UTC; request and readback began at 22:31:28.944, and readback completed at 22:31:29.159. Later identity checks did not change those timestamps. Personal review preceded the sole permanent game binding.

One manually chosen 120 ms Return gesture opened Game. Its independent UP acknowledged 130 ms after DOWN dispatch and the physical Return state was zero. The next game-menu capture created a ticket, then encountered a held paint before readback admission at 22:31:53.837. The error explicitly says post-ticket readiness changed: held/in-flight paint; no retry. It is an actual terminal capture failure, not a passive precheck. No later command or ticket was sent. original/first-failure.png is a personally reviewed raw diagnostic showing START GAME and the Game submenu; it was never promoted to an admitted scene or used for input. No round was started and no human cycle identified. Browser close was host teardown, not native Quit.

Native controller exit 1, Chrome exit 0, watchdog exit 0 (unfired), supervisor child exit 0 and outer caller exit 0 all have actual waits and both EOFs. The outer zero records custody, not game success. Runtime elapsed time was 67.330 seconds through the native wait. Recorded processes are absent, HTTP is closed, the profile is removed and the heavy lock is released. Before releasing source pins to thread-return-sentinel, all 677 files, all 92 media hashes and seven read-only source hashes matched. Selected images and hashes are named in the result. Box ID/expiry are null, state NOT_CREATED, stop receipt N/A.

The pre-ticket fix is verified and published; the subsequent post-ticket guard remains terminal. This original and all predecessors stay closed. The bounded task stops after this report; thread-return-sentinel is the next runtime task. No human win, full-27, FPS, audio, parity or physical-GPU claim is made.