GLTron personal handoff margin blocker — 2026-11-04

The fresh full-CLI offline demonstration failed at 6,824 ms from original capture start to the next reviewed measurement action. It did not meet the required 3,500 ms margin or the unchanged 5,000 ms guard. Zero Chrome launches occurred. No original gameplay interval, after-image review, WebGL backend observation or entered-update rate is claimed for this assignment.

Input: e942468008dfed383093b82ba800849b3efb7bf4 and gltron-low-latency-webgl-gameplay-20261104.md. Evidence: scratch/runs/20261104-gltron-reviewed-handoff-margin/. Accepted software evidence and historical qualification were reused without another loop.

Actual offline path

The current complete CLI ran from a fresh directory with freshly computed path/SHA pins, both compiled-module probes, the real HTTP server, command mailbox, atomic capture ACK producer and actual CLI client. The accepted run-specific browser fixture was loaded by reference, without copying its source. It supplies an explicit plain browser object; public child-process entry points and the independent native asynchronous spawn entry prohibit launches. Only exact read-only Git inventory calls are allowed. Its private ancestor lock and synthetic release board cannot authorize an actual browser.

The screenshot mock injected the historical 1,055 ms delay. The complete 800×600 lossless PNG, original host/page start clocks, callbacks, generation, hash and exact successful ACK remained intact. The accepted runtime source, ownership/liveness/native input/scene guards, original 16 ms pacing, first cause persistence, command blocking and automatic normal closure were unchanged.

The measured path removed optional shell login startup and extra ACK output: the actual --command CLI ran with login:false, its ACK was parsed within the tool invocation, and its exact capture path went directly to view_image with original detail. There was no image conversion, rescaling, retimestamping or additional model turn between ACK and image delivery. I personally viewed the complete alternating blue checkerboard. The immediately following tool invocation submitted the actual measurement command; fixture automation did not make the personal-review claim or submit that action.

Recorded boundary UTC / duration
Original capture start 21:48:20.092Z
PNG complete / successful ACK 21:48:21.148Z / 1,056 ms
ACK client complete / image helper call 21:48:21.158Z / another 10 ms
Image helper returned 21:48:21.164Z / another 6 ms
Next invocation after personal viewing 21:48:26.870Z / another 5,706 ms
Actual CLI prequeue guard rejected 21:48:26.916Z / another 46 ms
Capture start to rejected action 6,824 ms

The 5,706 ms interval includes image delivery, model processing and next-call handoff; the available clocks cannot attribute it more narrowly. The image helper's six milliseconds does not measure when the model finished seeing the image. The capture-to-next-invocation interval alone already exceeded the guard. Removing all 46 ms of subsequent command startup would not pass. No measured source defect explains a removable multi-second delay, so the FPS helpers and tests were left unchanged. Assuming a faster subsequent review would not demonstrate the requested margin.

First fault and closure

client-first-failure.json records queued:false and the 6,824 ms cause. No measurement command entered the mailbox and no mock counter armed. The unchanged owner detected that cause and requested normal close without another operator command. The full fixture exited zero, its HTTP port 46079 refused connections, and its private lock became available. The mock's chromePid is its fixture Node PID, not a Chrome process. Independent spawn prohibition recorded zero attempts and zero real Chrome launches.

The 81 unchanged focused checks and full locked build passed. Source evidence records the tested main SHA, relevant binary diff, new-note patch and hashes of the referenced fixture and modules. Media and modules remain at existing paths with SHA-256 references; no source/media/module archives were made.

This task stops at the concrete offline transport/personal-review blocker. The conditional runtime stage was never entered, so the new Stronghold assignment release was neither consumed nor replaced by an older release. A future runtime requires a real full-path demonstration at or below 3,500 ms, then the exact fresh stronghold-named-observation-gameplay-nov04 release dependency and separate coordinator authorization. No weakening of the five-second guard or automated personal review is a repair. Box ID/state/expiry are N/A; coordinator retains runtime lifecycle and shutdown ownership.