Recoil power-control objective: interrupted run recovery, 2026-11-08
Run: scratch/runs/20261108-recoil-power-control-objective/.
The force-field objective remains unproved. Recovery confirmed the entire original command ledger and unchanged live identity. A new passive capture shows a burning wreck under the native LOAD GAME / NEW GAME / QUIT menu. Continuing this mission from that view would require loading or restarting, outside this recovery's authorization. No recovery gameplay input was sent. The only new commands were passive capture 25 and orderly browser quit 26. No earlier command, including successful capture 24, was repeated.
The original worker's launcher handle 45742 ended with terminal status 143; its worker/Codex processes were absent and no outer final/result was present. The interruption cause is unknown. This note and the final result are recovery reports, not a clean uninterrupted first-attempt report. The owner reported that root sent no game input during the interruption.
The native Mission 1 brief, personally reviewed again in
original/mission-brief.png, says the beach force field blocks the hilltop
enemy base: find and destroy its power control. The original worker opened
the ordinary R transmission, O objectives and M map before steering.
original/objective.png and original/map.png show objectives 0/4,
enemies killed 0/102 and score 0. The original controls then turn around the
wreck, advance, turn inland and approach the cliff. The labels “approach-road”
and “road-heading” are command labels, not proof of reaching a road or new area.
original/road-heading-settled.png still faces the cliff, with score 0 and
health 47. No power-control installation was identified or attacked, and no
counter change, field deactivation or newly accessible area was established.
The previously accepted biped kill/1550 score belongs to the earlier run.
At takeover, recovery-audit.json records 25 contiguous commands 0–24 with
exact successful ACK IDs, no refusal, no missing reply, no writer lock and no
browser failure or sticky transport closure. The browser and passive monitor
were still live. Its last two monitor samples show all arrow keys UP, buttons
zero, and the same page Canvas2D renderer object. The existing detached
normal/watchdog delegates and supervisor were alive with the original deadlines;
none was replaced or extended.
New capture original/recovery-passive-25.png was personally reviewed.
Its capture chain independently recovers the live module and confirms all
16 identity fields against command 24: document epoch
b9e34a51-72ba-4a98-b990-9064e5527f8d, process 1000/endpoint 1/slot 0,
the same process/link/Worker/module object tokens, Worker target
2D7E820836E0D9253E530DEC4F6250E6, and instance
4ec270b7-e9e0-4e47-b0b7-7e0645207bf0:1.
There is no owning D3D render-worker endpoint. The passive monitor's renderer
object is epoch:5. Guest slice sequence advances 30510→43082, which confirms
execution activity, not simulation steps or FPS. The native menu and wreck are
consistent with player destruction while unattended; the exact transition and
cause were not captured. No emulator defect or objective consequence is inferred.
Earlier failures remain intact and are distinguished from input uncertainty:
launch-remote-result.json: a prelaunch pin guard found missingzbd/soundsM.zbd, before any browser was spawned. The provision ledger then records that exact file; the guarded launch at 02:20:35Z verifies all 726 pins and records one original browser launch.fetch-original-startup-remote-result.json: the premature read oforiginal/startup.pngfailed because that file did not yet exist. The later ready file and captures confirm startup completed. This was an artifact retrieval error, not guest failure, input UNKNOWN or command replay.recovery-terminal-remote-result.json: the first terminal inventory's seven-second process-absence guard failed during shutdown. The separate read-onlyrecovery-cleanup-diagnose.jsonsubsequently finds every recorded PID absent and both ports refused.recovery-guarded-inventory.pychecks ready/cleanup/file guards, exact ACKs, pins and process absence before producing the successful terminal inventory. The failed record is retained.
The legacy cleanup.json includes the generic text “UNKNOWN; never replay”
in its pending field. The authoritative numbered ledger has all 27 exact
successful ACKs, including recovery capture and quit, with no outstanding
command. That boilerplate is not evidence of an unacknowledged game action.
The original fixture-identity.json reports zero registered matches and is
not used as fixture proof: tested-source-identity.json and the supplied
108 fixture pins provide the independently checked identity instead.
Tested source: main 599c695c374e796b8b07df168be6a10d7db125b5, retained
relevant-source.patch and new-lib_wininit-boot.js.patch. All 726 runtime
pins match locally and after the remote run; all 108 supplied fixture pins
(107 original files plus manifest) match. Original executable SHA-256:
82632b1b46153014af0caf1e0c9dfdebbb0d8482ba65b2735064f10972afa076.
The actual CDP-recovered Worker module matches the staged build:
f9650607181301b5d2a9990420f437bd9f1ca10d40ac73067339f07499d53121.
Git plus patches is the source archive. No source tree, repeated game media or
test binaries were archived into the run. Recovery inputs and this note are
also retained as patches.
Validation: verify-recovery.cjs passes the complete ledger, live identity,
module, fixture/runtime hashes, all 101 remote PNG CRC/IDAT decodes and
capture hashes, and cleanup/lifecycle receipts. The original local HTTP staging
check passes for 113 URLs, and original native key/type/combo preflights pass.
Recovery changes only this documentation and ignored evidence/helpers; no
emulator repair or new locked-build claim is made. The environment is headless
Chrome 151.0.7922.108 / V8 15.1.206.16, 1024×768, cross-origin isolated,
Threads enabled, unchanged registered Software route. Software Canvas2D is
not GL proof; audio, FPS and full playability remain unverified.
Owned evidence references and note whitespace checks pass. The global evidence
checker retains seven unrelated containment errors in other runs.
All 393 remote artifacts / 34,643,824 bytes were retrieved and independently
SHA-256 verified before requesting the existing normal-stop delegate.
Unique captures, held frames, logs and actual module bytes are retained.
The browser exited 0, all 15 recorded PIDs disappeared, HTTP/CDP ports refused
connections, profile/writer lock were absent, and keys/buttons were released.
Both original lifecycle children exited 0 and the supervisor ended empty.
Box bx_n8ujzs67 is archived, stopped at 02:29:25.020Z,
expiry/archive-after null (original expiry 02:45:56.336Z).
Stop stop_a16986fbed7f1006 completed at 02:29:25.039Z,
forced=false; final-stop-receipt.json retains the final GET.
The original browser deadline was 02:41:56.164Z and watchdog deadline
02:44:56.502Z. Recovery sampled local free disk above 5 GB; no RAM gate,
new box, resume, deadline extension, delete, public deployment or nested worker
was used.
Next work requires a separately authorized mission attempt or supported native checkpoint. It should navigate toward the reviewed power-control objective with frequent personal review and bounded controls. This interrupted run cannot establish the force-field consequence, and remains stopped.