Penguin Command: bounded cannon block-entry decision trace, 2026-10-24
Historical source last updated, 2026-10-24. Runtime and implementation claims below describe that dated source; see current app notes and source for subsequent changes.
The original A action passes all observed early cannon rejection tests and begins initializing shot record zero. Completed acceptance remains unproven: the sole authorized trace exhausted its 524,288-byte output allowance in the initialization loop before count/ammo writes were retained. No exact rejecting branch was witnessed. No projectile, score, wave, or general gameplay claim is made.
The witnessed block path is 406d80 → 406d9c → 406da5 → 406db2 → 406ddf → 406dee → 406df7 → 406e28 → 406eb0, then repeated 406eb0. At 406d9c, lazy subtraction operands are target Y177 minus510; at 406da5, EDI1 minus3 confirms the invalid-cannon jg 406f00 was not taken. Reverse at4130a0 is0, so jne 406f08 was not taken. Active count at4130a4 is0; the duplicate-target test is bypassed by je 406ddf. Ammo is10, so jle 406f00 is not taken. Count0 minus19 at406df7 confirms jg 406f16 was not taken. Registers, lazy flags and exact memory bytes are in raw output.log; supporting original PE disassembly is retained beside it.
At first406eb0, record4d0750 changes from32zero bytes to 0000484200c000448da11e3e9c5c8fbe00000000000000000000000000000000: floats50,515 and two velocity values. EDX points at4d0750. Count remains0 and ammo10/10/10 throughout the complete retained hits. This proves partial shot initialization, not the subsequent count increment406ece or ammo store406eec. There are1005 logged loop entries (the final line is partial). The file stops at exactly524288 bytes; natural supervisor returncode0 cannot establish full diagnostic completion after the output limit. The finite loop's detailed stream exceeds the allowance. No second execution, repair, rebuild, new instrumentation or forced game state was performed.
The exact predecessor menu/aim/A schedule is retained in decision-args.json: menu down34000/up34500, aim34800, A down40000/up40200. The raw log records scheduled A down40000 and cannon entry with EDX61 and prior compare61−61=0. The actual A dequeue/DispatchMessage batch and key-up delivery are unavailable in this capture: those messages use the runner's batch-buffered log while EIP lines stream immediately; the file fills first. Predecessor batch40033 is not substituted. This is an additional acceptance limitation.
The first checkpoint verified existing --trace-eip-range=0x406d80-0x406f50, --trace-eip-from=39999, detail, stream and five dumps8/12/4/4/32 against runner main5691793a58b4670fd2a621114d7926100a1e47de and historical sourceb41a7c5f4b752d1f0bdab1141a11d4fed4dc6f9a. Exact source slices, blob/file/excerpt identities and supporting imports/exports from the accepted retained contract are in static-checkpoint.json. Source dispatch guards disable chained execution while tracing and the historical run loop calls host.log_eip at block entry. The historical module is8fac39936b6df2418ad86cab118cef19a65a727f649fbc14c412b37818904686; runner is3c59fd034391e069f24a6047cf96cd8dac84a791d7d7591d55e5f7cb6e161a5e; original PE is14d6e430c26c44c94ef5c176eca2f66edd82dd66fd155d29a5596e3d6e170fc1. This is not current-main Wasm. Fresh preflight rehashed all291 existing staged files15760340 bytes without restaging, admitted7365206016 bytes available memory against2684354560 minimum, and found no runner process before launch.
Exactly one guest executed using the existing supervisor, with task output paths and strict file limit,512MiB arena/oldspace,90CPU/wall seconds,60guest seconds,43000batches,200ms/batch,840×660, no-build/quiet flags/no-close. It ran1.660seconds; supervisor reports no remaining process group. command.json and terminal.json retain exact argv, PID, timestamps and limits. The output limit prevented retaining normal batch-cap/exit summary, so no termination-cause inference is made from missing logs. No screenshot was needed.