SuperTux 0320: setup capture requested before readiness

The requested new coin/block consequence or onward exit was not achieved. The one authorized original stopped at its first helper UNKNOWN. I submitted the setup menu capture before the driver had published readiness. Its unchanged 30-second ACK deadline expired at 03:22:26.108Z. This was my sequencing error; no emulator gameplay defect was established. No code changed.

Evidence: scratch/runs/20261109-supertux-client-capture-onward-0320/. The numbered command inventory contains only menu, action capture, with no consumed marker; native-input.json is empty. Native Pause was never reached, and no gameplay action or review-client capture/choice ran. There was no retry or subsequent worker input/capture request. However, the already-running driver automatically made startup.png at 03:22:33.420Z, 7.312 seconds after the helper failure, before its consumer observed sticky transport closure at 03:22:33.687Z. Thus the strict no-later-capture sequence was not fully satisfied. I personally inspected that recovered PNG: black loading viewport, small white patch, and launch log; no native menu, Pause, or gameplay consequence. Its SHA-256 is 294fcf5bef67deaf3460ec60ea8a779e2578c21d08df8f053c985e6f96d7717a.

Before allocation, a new tiny offline check exercised the production review client, shipped page in a VM, and real HTTP. It verified that a separate setup capture leaves this client's current empty and choices disabled, then used the helper's explicit capture expression and reviewed choice expression. The resulting command order was capture, then pause-step, with the exact client capture reference. The 12-second bound stayed unchanged; no local browser/game or route/detector/replay/retimestamp/freshness bypass was used. This proves the offline client sequence, not actual native gameplay. offline-initial-sequence.json.gz records it. Prior prepared checks and locked build, including df0932c1c failure replies and airborne-only neutral settling, were reopened and verified by exact SHA/size; identical tests were not rerun.

The exact Q2upstream0301 release was 03:16:04Z, box bx_m6w3a76f, receipt stop_c41c040c0474c08a. Providers were pinned to published Git 0060f6935fbed715655c59cf4efa28b5141ee963, including custody helper SHA fb670a47d4712aa44fd2fae3d2047587374c0e1c2bb4e927ff87572147dfcfff; concurrent offline edits were excluded. The new final manifest covers 1,512 providers / 61,533,396 bytes, SHA 29a816e03799bec4abc226c2ed33a4672d83816158a30f0b42ece9e5dc04436e. CPU remained a320d4c85156644fd12b6134d66b11784f3100b9b135756206c087cdacd43295; native observation stayed OFF and existing caps stayed unchanged. Git commits and small input patches preserve source; no source/media/module archive was made.

Independent watchdog UUID e1f56dec-d15f-4b84-86b4-e5d5d4b2c483, PID 936115/start224166487, was observable at 03:20:55.670Z before finite-TTL creation. Its immutable deadline was 03:45:55.555Z. The sole box bx_nv58ujct was created at 03:21:08.191Z, TTL 1,487 s, initial archiveAfter=03:45:55.052Z; expiresAt was not returned. The original execution began at 03:21:43.144Z.

Automatic no-input cleanup received Browser.close ACK but required TERM/KILL: normal=false, 26,008 ms under the unchanged 30-second cap, terminal code 0 and drained streams. This is not normal browser-close acceptance. Generic cleanup retained pending="UNKNOWN; never replay"; both physical drain records separately report 0. Kernel custody verified 181 owned identities / 181 actual reaps / 1,639 events, including before-wait witnesses, complete=true and no unresolved lineage. Kernel parent 6697/start17241 actually waited 1; outer parent 6694/start17231 waited 0; both have separate stdout/stderr EOF. The client exited 0 with both EOFs. Ports 36515, 42095 and 8787, writable run FDs, profile and locks drained. Historical CLONE_UNTRACED/nonleader-exec/external-signal limitations remain UNKNOWN.

Recovery retained 74 regular files / 55 unique transports / 412,572 stored bytes. Original/stored SHA-256, sizes, independent reopen, GNU gzip CRC and exact-set checks passed before delegated normal archive. Box archive completed at 03:23:37.416Z, receipt stop_c8259d6d99763a57, forced=false, error=null, stoppedAt=03:23:37.398Z, final archiveAfter=null. Local runtime and watchdog-supervisor wrappers actually waited 0 with separate EOFs; watchdog waited 0 and board tail -15, both streams closed. Recorded local identities were absent. Physical release was posted at 03:24:16Z, with the automatic-capture correction at 03:24:34Z.

Free /home remained above 1.5 GB. No agents/workers, peer boxes, local browser, heavy local execution, RAM gate or deployment. Full 27/parity/audio/FPS remain open. A future separately authorized original must wait for actual readiness before its first setup request, then reach Pause, explicitly request the review client's initial capture, inspect it, and choose promptly. This single task stops.