Pengo M04:301,030 verified /300k PASS /0 MODEL CALLS
301,030 alive /301,030 at tape end. M04 A10 passes its frozen300,000 target.
301,030 alive / 301,030 at tape end; four lives remain, zero detected deaths. 52 chronological rounds completed; ACT 5 of cycle 4 remains unfinished with four objectives.
TRAJECTORY SEARCH / PRIVILEGED / REAL TIME REPLAY VERIFICATION. 0 MODEL CALLS. This is a verified trajectory prefix from one canonical reset. Robustness across different starts is not established. Human competition, privileged trajectory search and an autonomous agent remain separate categories.
Independent admission
Two NEW complete native reset executions replay all167,028 physical frames, 0 LOAD /0 RAM writes. Both agree on actual input, the whole declared trace and sampled rendered pixels/exposed devices. The independent terminal-pixel gate reads301,030 in BOTH proofs and rejects corrupted glyphs. Pixel and chronological-round receipts are bound to the immutable A10 result hash.
The chronological audit counts three full16-ACT cycles and ACT1–4 of cycle4: 52 completed rounds. ACT5 of cycle4 is partial, with four objectives. Four lives remain and zero deaths are detected under the frozen scoped death guard. Three initial lives and the observed bonus life are retained; raw nonzero phase bytes are not treated as death by themselves.
A THIRD NEW complete native reset records the admitted input, with no extension, LOAD or RAM writes. Whole trace, input and sampled pixels/devices match BOTH new proofs. Every second native frame is shown: 83,514 frames at30.303030fps, full duration 2755.962s (45m55.962s). No intentional cuts or speed-up. Full decode and text-bound checks pass; ten actual composited previews and three frames decoded from the MP4, including its terminal frame, were visually inspected. This sampling is not an assertion that each video frame was separately visually inspected.
The nominal native game rate is60.606Hz. Native proof operation walls are 2755.964873s and 2755.952501s. The new capture/comparison operation takes 2910.016s, which includes screenshot I/O and is longer than the nominal emulated duration. Encoding takes 2040.203s of software time. These are different measurements; the video’s emulated cadence does not establish uninterrupted host frame pacing.
Search and exact cost scopes
Search uses validated healthy-PLAY restore, CPU beam search, logical-cell macros and witnessed rollback. A10’s immutable configuration uses at most16 workers, base6/20 search and adaptive12/40 then24/80, calibrated per-ACT watchdogs and delayed-removal guards. Death, UNKNOWN, disagreement and raw17 restore exclusions remain. Final admission and this recording use complete native reset processes. 0 MODEL CALLS means no model decision calls in this execution.
| Scope | Physical frame evaluations | MAME processes | Wall seconds |
|---|---|---|---|
| A10 search and BOTH final proofs, closed exact | 5,243,483 | 47,482 | 9334.282 |
| Additional whole M04 recording/comparison, closed exact | 167,028 | 1 | 2910.016 |
| Encoding, QA and publication, software only | 0 | 0 | 2040.203 encoding |
A10 records48 planning steps,735 commits,2 rollbacks,0 recovery returns and8 handoffs. Decisions and rollback counts remain separate. Score yield is 109.229points per nominal native second for the complete tape; search wall time is not substituted into that gameplay pace.
Earlier and active campaign costs are excluded from these closed scopes, rather than represented as zero. The dated committed hourly snapshot at 2026-10-07T18:00:19.691684+00:00 contains 193 registered identifiers and a known lower bound of 110,911,107 physical frame evaluations /232,458 owned launches. M05 is running in that snapshot, so total cost is provisional. Reused verification corpora and historical categories make per-identifier rows non-additive. Dated outcomes and accounting scope.
Native verification · Pixel gate · Chronological audit · Whole recording evidence · Video metadata · Visual/decode checks · Capture/publication protocol.
Input SHA256: 0814d56d3f1811cc9ebfc8c2253206489bcc1db369bf14fcdef856b00194b176.
Observed trace SHA256: 42c815747bb5ca380cfcaf731c90fc351ab6840f37b625ac50f383d102523597.
Video SHA256: e73e12393fe3569b6f79fdc68825f1efadce53dd8e45398ce895333876e657c1.
Declared trace scope: PENGO01 header, physical frame indices,4096 bytes of RAM0x8000–0x8fff,16 sprite coordinates and five driver flags. Actual input, sampled pixels and exposed device state are checked separately. Other CPU, scheduler, timer and device state are outside this hash; complete-machine identity remains UNKNOWN. No ROM, raw RAM dump, save-state, game-code bytes, input tape or model weights are distributed. Video phase labels come from the frozen runtime reader; unclassified phases remain UNKNOWN. This presentation introduces no new gameplay reader.
Audit and next milestones
The W1/W2/X1/X2 audit is complete. Removing ordinary30-point rewards changes0/13 ordering comparisons; Stefano withdrew the proposed ordering/reward changes. A10 ordering and the30 reward remain. The correctedX1 gain is 1,890 =2×400 +2×500 +3×30; ice disappearance alone does not establish a Pengo block break. Retained audit · Stefano’s superseding instructions.
M05 targets500,000 with A10 configuration unchanged. Its running observations are not part of this admitted result. Persistent MAME experiments proceed separately with new chained-restore fidelity and cannot alter the frozen M05 run. E072v1 remains INCOMPLETE for a CRLF request-transport fault; its20 fresh controls passed, and459 frames/22 processes remain in the ledger. No persistent restore was admitted from that failure. Diamond probes resume only after the required persistent fidelity and the retained prior NUL explanation.
After500k, the first legal 655,360 crossing is a separate preregistered milestone: word and pixel evidence in BOTH new full native proofs, modular positive awards,1/2/3-digit HUD checks and observed epochs. Cumulative certified points remain separate from HUD. Legal laboratory rollover has not been observed. Only after that contract comes1,000,000 cumulative; ultimate numeric goal 1,492,230.
Settings and human comparison
MAME0.289 /pengoja /Medium /Upright /THREE initial lives /bonus30,000 /Rack Test OFF /DSW1=176 /DSW2=204. The supplied human benchmark is 1,492,220, listed for OOO by Twin Galaxies, under FIVE-initial-life Points rules. Human1,492,220 and our target1,492,230 are distinct. Different settings and categories establish no human rank, world record, full-game completion or autonomous-agent result. Score-target provenance.
Project Pengo · M01 /21,210 · M02 /31,040 · M03 /100,400 · M03B /216,480 · Complete YouTube replay. Historical Pengo and Pleiads evidence is preserved.