GLTron locked live interval — 2026-11-04
Run: scratch/runs/20261104-gltron-locked-live-interval/.
Input: 58273d2bfb18d2ce7e5bd10c7941c2f628caee4e and
gltron-complete-handoff-interval-20261104.md.
Software completed an actual original logical-update interval: 50 entered nonzero-dt updates over 5,553.089999914169 monotonic milliseconds, or 9.00399597355217 updates/s. Personally VIEWed the exact active-before and after-measure captures relayed from successful actual command ACKs. Both show the living yellow player and its trail. The after image says PLAYER 4 CRASHED; the observed player is player one, still speed 8.5, winner -1, pauseflag 1, with unchanged game 4444832 and player-data 11124300. This is a software-only observation. WebGL never launched, so there is no paired comparison.
The subsequent first fault was player-one death during continued ordinary
steering, at 20:39:25.249Z, after the interval's after capture at
20:39:19.732Z. The exact first-fault speed is -1. The entire pair stopped;
the attempted next command was rejected locally as runtime closed and was
never queued. No original or backend retry followed.
Before Chrome, the actual historical replay traversed command queue, shot
producer, successful ACK and exact path/SHA/epoch validation. Personally
VIEWed that exact ACK path in the capture invocation: GLTron logo, yellow
cycle, Game/Video/Audio/Quit menu. The immediately following actual
review-only passed at 3,856 ms, and the actual pre-Chrome consumer accepted it.
This rehearsal is historical only. The old 4,521 ms proof was not reused.
The genuine anchored MDK sender release at 20:29:22Z was consumed. One
external /tmp/wine-local-runtime.lock covered the entire actual run; the
runner recorded its inherited ancestor PID/descriptor before server/Chrome.
Personally VIEWed actual startup, main menu, Game menu and paused arena before
dependent ordinary inputs. The first startup capture showed the black loading
canvas; a later main-menu capture established readiness before Enter. The
active-before capture and personal VIEW were in the same ACK invocation.
The immediate measurement armed at review age 4,686.990000009537 ms, below
the unchanged 5,000 ms guard. Original marker 0x4051a9 counted lifetime
entries with pacing zero, canonical module
78a0b1d61b23428a1ae909d739440ae0de059bbbd84a69a2ad3e4c37471ad664,
owning main thread 1, generation 1, translated callback offset 250416,
chain 0/uop 1 and tier guard zero. Monotonic read brackets, minimum five
seconds, reset/wrap, foreign counters, visibility, frozen/host steps,
scene/liveness and data identity all passed. No wraps occurred. Callback
0x42a200 alone was never used as proof of gameplay. Balanced trusted native
A/S steering and marker/observer overhead are included. No guest code, state,
configuration or pacing writes, freeze, host steps or time exclusions.
Physical GPU performance remains unproved.
Offline causal repairs after the first fault
The old steering loop imposed 300 ms after each completed native turn. The retained last live sample at 20:39:25.121Z had already acknowledged the turn from direction 1/trail 6 to direction 2/trail 7. Only 5.230932235717773 units remained ahead; side clearances were 86.03533935546875 and 280.76788330078125. The last trusted S keyup was at page time 44919.09999990463; this sample at 45192.669999837875 was only 273.5699999332428 ms later. The old cooldown suppressed the required S turn.
The repaired actual loop waits for the original direction/trail acknowledgement
before sending another balanced native turn. It no longer suppresses an
acknowledged narrow-corridor escape with a fixed wall-clock cooldown.
steering-causal-replay.json executes both actual loop bodies with the same
full retained sample: old producer emits no input; repaired producer emits S.
Focused tests also require pending-turn suppression and unchanged liveness
rejection. This proves the decision repair offline, not future native delivery
timing or game survival. There was no post-repair Chrome.
The release floor now matches the genuine 20:29:22Z prerequisite rather than the predecessor's 20:19:19Z floor. Actual old/new parser replay demonstrates that an earlier same-sender line formerly passed and now rejects. Anchored sender, waiting/foreign/old/future rejection remain required. The actual unflocked CLI entry point is additionally tested with all child-process, server and Chrome paths trapped; a missing guard can fail the test but cannot launch a browser. Failed ACK/path/hash/epoch/stale-proof negatives remain.
Validation: 76 focused checks, including both unchanged compiled modules executing original branches under all four tiers and scene/liveness negatives; full locked build, syntax, diff and test-tier checks. Pre-Chrome preparation retains 654 pins, 92 media, 365 prior HTTP requests and 1,010 offline routes; actual served responses are checked against those pins. Existing 50 marker checks/module remain reusable because both module hashes are unchanged.
Normal Browser.close: Chrome PID 1107211/start 187192343 actually exited 0/signal null at 20:39:25.592Z. Node PID 1107185/start 187192297 and wrapper exited 0. Exact PIDs and owned-profile processes are absent, HTTP 35805 and DevTools 40929 refuse connections, server closed, flock available and profile removed after closure checks. Browser lifetime 51,412 ms; no foreign kills. Box ID/state/expiry N/A; coordinator owns future boxes and shutdown.
The prior 82-file/1,539,457-byte and 103-file/3,854,998-byte seals remain unchanged. Exact Git SHA plus relevant binary diff and owned-new-note patch are the source archive. Media and modules are referenced by existing paths and SHA-256 only; no source/media/module snapshots or tar archives.
Next: any future original attempt requires separate authorization, a fresh complete rehearsal, current release and inherited flock, and all unchanged live interval/personal after-review guards. This assignment stops after publishing the bounded software observation and offline repairs.
Interrupted-worker recovery
The original worker terminated without a final result. Recovery run
scratch/runs/20261104-gltron-land-live-observation/ adopted its released
implementation and note, inspected the retained before/after images and raw
counter evidence, and reran the 76 focused checks and locked full build.
The CLI mutation check additionally requires the guardless producer to reach
the missing handoff proof; the guarded producer must reject on the missing
flock before reading that proof. Every child-process/server/Chrome entry in
this CLI fixture is trapped. Recovery launched no Chrome, original game or
heavy runtime and did not alter the earlier run or seals.
The personal live reviews above belong to the original worker. Recovery's image inspection is retrospective and does not claim a second live review or a new measurement. The measured rate counts entered nonzero-dt updates, not completed updates, display frames, or physical GPU performance. The offline steering repair has no post-repair live survival result.