SuperTux onward attempt: review-client capture handoff missed

The requested new coin, interactive question-block consequence, or level exit was not achieved. The sole fresh original reached Welcome to Antarctica and native Pause. I personally inspected all twelve setup images. I then attempted the first right-100 through the review page without first requesting that client's own initial capture. Its choices were correctly disabled: createClient starts with current=null; a capture made through the separate setup transport does not supply the client's capture token. This was my command-sequencing mistake, not an established emulator/runtime defect. No production code changed.

Evidence: scratch/runs/20261109-supertux-interactive-onward-0300/. personally-inspected-frames.json.gz names the exact images, original times and SHA-256 values; the lossless originals are in review/*.png.gz and the hash-addressed remote recovery set. initial-pause at 03:07:49.419Z shows Tux grounded on the snowy floor with native Pause visible. Score 50 and coins 2 were already present at level entry; neither is a new consequence of this run. The first gameplay helper began about 5,943 ms after that setup capture and failed at 03:07:55Z with review choice disabled, before click(), HTTP submission, numbered game command, or physical gameplay input. This was not a stale-frame refusal. The helper's newly opened review page had no client capture. failure-command-audit.json.gz verifies zero accepted Pause steps and zero review-client capture commands. No later game input or capture followed the failure. Only ordinary no-input cleanup quit followed. There was no retry, route, detector, replay, retimestamp, freshness workaround, cap change, or new scenery/collision claim. The accepted first-enemy stomp and floor landings remain prior evidence; the early snowy pillars remain scenery.

Offline preparation reused 0251 by exact SHA/size, including published df0932c1c failure replies and the explicit airborne=true, grounded=false, empty-key 200 ms neutral-settle helper. Neither airborne settling nor any gameplay choice ran here. Exact singular Enter/ArrowDown/Escape composed validation and accepted unchanged focused/build evidence were hash-verified and reused, not rerun. offline-reuse.json.gz and repair-build-reuse-verified.json.gz record the paths and hashes. The final local/remote barrier matched 1,497 providers / 61,507,039 bytes, manifest SHA-256 4fe8f43e8291ebb7a1889aa3e4067472de013254f335db06f78c2bfdfba619b4. Preparation selected current published providers at Git 0060f6935fbed715655c59cf4efa28b5141ee963; the CPU stayed a320d4c85156644fd12b6134d66b11784f3100b9b135756206c087cdacd43295. Native observation remained OFF. Small run-input patches plus Git preserve source recovery; no source/media/module archive was made.

The exact Warcraft0244 release for bx_6psgpzsk was posted at 03:03:57Z. The board watcher actually waited 0 and awaited its tail, recording PID/start and separate stdout/stderr EOF. The independent watchdog was prearmed at 03:04:53Z, UUID 28031804-57f9-4541-bf41-75f1bcf9361c, PID 913454/start224070253, immutable deadline 03:29:53.176Z. Only then was bx_bw4vg5br created at 03:04:59.817Z, TTL 1,493 s, initial archiveAfter=03:29:52.671Z; expiresAt was not returned. The original began at 03:05:45.177Z. No second original was started.

Browser.close received ACK and completed normally in 11,376 ms under the unchanged 30-second cap: terminal code 0, signal null, streams drained, no escalation. The review client exited 0 with both EOFs. Both physical-drain records report pending 0. The generic cleanup record still says pending="UNKNOWN; never replay"; that field is preserved, not rewritten. Custody verified 249 owned identities / 249 actual reaps / 2,434 events, including before-wait PID/start witnesses. Kernel parent 6345/start8843 and outer parent 6339/start8832 actually waited 0 with separate stdout and stderr EOF. Ports 36007, 42155 and 8787, writable run FDs, active profile, input/client locks and runtime lock drained. Historical CLONE_UNTRACED, nonleader-exec and external-signal limitations remain UNKNOWN.

Recovery accounted for 214 retained regular files, including pinned input references, through 169 unique transports / 3,074,569 stored bytes. Original/stored SHA-256, sizes, independent reopen, GNU gzip CRC and exact-set checks passed before delegated normal stop. The inventory's inactive Chrome-cache exclusion selected zero files; unique game evidence was retained. The box was archived at 03:09:01.171Z; stop receipt stop_063972e579437d2e completed 03:09:01.194Z, forced=false, error=null, final archiveAfter=null, expiresAt not returned. The local runtime/Boat launch and watchdog-supervisor wrappers actually waited 0 with separate stdout/stderr EOF. Watchdog waited 0; its tail 913455/start224070253 waited -15 with both streams closed. Runtime release was posted at 03:09:17Z.

Free /home stayed above 1.5 GB at recorded allocation/recovery checks after coordinator-restored headroom. No local browser, local heavy execution, peer box, agent/worker, RAM gate or deployment was used. Next work requires a separately authorized fresh original: after setup reaches native Pause, request and personally inspect the review client's initial fresh capture, then submit the chosen bounded action promptly under the unchanged 12-second bound. This run stops here under the first-owned-failure rule.