Pengo100k milestone: 100,400 verified

Best verified planner trajectory: 100,400 alive / 100,400 at tape end. 4 lives remain and no death is detected in the retained 45,789-frame tape. 15 chronological rounds are complete; final ACT label 16 is unfinished with 9 objectives. ACT labels wrap after16; completion is counted chronologically.

Category: TRAJECTORY SEARCH / PRIVILEGED / REAL TIME REPLAY VERIFICATION. Two NEW native-speed complete reset processes match exact actual input, the whole declared prefix trace, each other and sampled rendered pixels/exposed CPU-driver registers;0LOAD and0RAM writes. This is not a full-game/world record or an autonomous agent. Comparable leaderboard provenance remains unverified.

Independent promotion after retained null attempts

The100k sequence continues the earlier31,040verified prefix using validated healthy-PLAY restore search, cell macros and witnessed backtracking. Intermediate ACT handoffs use paired accelerated NEW reset processes as search evidence. Final promotion uses two NEW complete native reset replays, with speed explicitly requested and confirmed by manifests. No transitional state is restored. Native replay duration is 755.535seconds in the video at 30.303030fps. Chronological round instances: [{"instance":1,"ACT":1,"frame":2044,"score":1850},{"instance":2,"ACT":2,"frame":4492,"score":6000},{"instance":3,"ACT":3,"frame":9709,"score":22410},{"instance":4,"ACT":4,"frame":11721,"score":26040},{"instance":5,"ACT":5,"frame":14378,"score":35230},{"instance":6,"ACT":6,"frame":16616,"score":40780},{"instance":7,"ACT":7,"frame":19596,"score":46800},{"instance":8,"ACT":8,"frame":22345,"score":53090},{"instance":9,"ACT":9,"frame":25237,"score":59080},{"instance":10,"ACT":10,"frame":27799,"score":65300},{"instance":11,"ACT":11,"frame":31564,"score":71600},{"instance":12,"ACT":12,"frame":34839,"score":78420},{"instance":13,"ACT":13,"frame":38239,"score":84350},{"instance":14,"ACT":14,"frame":41162,"score":90510},{"instance":15,"ACT":15,"frame":44365,"score":97100}].

M03 attempts01,04 and05 remain NUL; attempts02 and03 remain INCOMPLETE. The original classifications and frozen artifacts remain intact. Attempt05 stopped on its40-decision watchdog at95,070observed points in ACT15, not a promoted score. Its continuation exposed a reader limitation, so scoring stopped before diagnosis.

The witnessed ACT15 sequence has a100-point contact with a stunned Sno-Bee at43,289, plus a separately confirmed ice crush. At43,456 two actors retire and two eggs hatch; the objective counter falls7to5 with no score change at that acknowledgement. The new prefix-only reader credits the causal contact once and later reconciles the combined retirement/hatching bijection. The extension is scoped to witnessed ACT1/2/15 contexts. Unsupported contexts remain UNKNOWN; death/life/state gates are preserved.

E051v1 mistakenly inherited accelerated=True in its supposed native launcher; it remains NUL_SPEED_PROTOCOL_VIOLATION, with18MAME/572,391frames counted. Its raw result is preserved with a separate authoritative adjudication. E051v1.1 corrects the launcher, then12NEW native replicas of neutral input and every legal macro plus600neutral frames pass paired trace/input/sampled pixels/devices, zero loads/writes and speed-manifest checks;18MAME/572,391frames. No score is promoted from either diagnostic.

E052candidate-check-v0 is NUL software because its comparison omitted the legacy event’s later censorship field. The revised criterion permits clearing only censorship strictly later than an independently witnessed retirement; death/life and all other state fields must remain identical. Softwarev0.1 passes without admitting the reader. The actual E052v1.2 admission then passes163frozen traces, including12correct native witnesses, zero unexpected differences, clone consistency and19positive/corrupted-context checks. It adds zero emulator frames. This is explicitly post-hoc scoped reader admission, frozen before the new scoring configuration.

Actual attempt06v1.2 starts at healthy43,453/95,070observed with the admitted reader and unchanged score policy. Its budget is45minutes,50million physical frames,20CPUworkers,width8/depth6,40decision progress watchdog and at most16accelerated paired handoffs. Every failed or diagnostic attempt remains counted. Scoring protocol, operations, reader protocol, reader admission, native diagnostic.

Declared evidence and limits

Observed trace SHA256 52426a655ce089210097a2085e33c87a001a201a49433af20895dae36da121b7 covers PENGO01header, physical frame indices,4,096bytes RAM0x8000–0x8fff,16sprite-coordinate bytes and5driver flags. Input,pixels,CPU,scheduler,timers and other devices are excluded from that hash. Actual input SHA256 a73add1ea978e571ec1c26cf616e3b2c0754cd8cec03e5c86074e1a5f8238cff is checked separately; rendered pixels and exposed CPU/driver registers are compared where sampled. Complete-machine identity remains UNKNOWN. No raw RAM dump,save-state,ROM,model weights or secrets are published. Video SHA256 5bd256938151f4837295a4539a550ef2c25dfca95937045e4253e0119ee8b9a0.

Full canonical input tape, registration, sanitized verification, video metadata, visual validation.

Fixed cumulative milestone cost: 87,176,655known physical frame evaluations and 54,921owned Pengo MAME launches. All counted diagnostics,NUL,INCOMPLETE and preflight configurations are included; reused captures and Pleiads work are excluded. Totals are shown fixed from video start. Complete cost snapshot.

Baseline remains pengoja/MAME0.289,3initial lives,Medium,Upright,bonus life30k,RackOff,DSW1=176/DSW2=204, native60.606Hz. Earned lives are reported as observed. Rack Test is diagnostic only and supplies no result seed. The earlier measured ACT3interior10,000diamond award remains scoped; no unverified wall/cross-act/repeated award is added to the policy. Two subsequent ACT14 constructive diamond probes remain NUL and do not change that award scope. E050v1.1 costs161,918frames/14MAME: its first ML correctly transitions(8,11)to(7,11), but a full-board comparison mistakes nonlocal ICE animations for failed action settling, then waits until a detected death. E050v1.2 costs162,687frames/55MAME: exact local MOVE/PUSH comparison and live-board rereading pass seven software regression guards. Four timing offsets reach15,15,17and4actions respectively; two stop at illegal next inputs and two at detected death. No alignment or award is observed. These open-loop geometric recipes are not a threat-aware strategy, and neither probe promotes a score. All costs remain included.

Next numerical target300,000, with a separate short pre-registration, measured whole-tape native reserve and hard07:00UTC deadline. Models/student,EmbeddingGemma,Qwen,PPO,new games and speed-record work remain suspended. Frozen PleiadsE007/E008 and historical Pengo artifacts remain unchanged. Previous milestones: 21,210 and 31,040.

Review update7October2026 / editorial v2.1

The verified tape and scientific result above remain unchanged. The corrected video names the validated restore search engine separately from final verification0LOAD/0RAMwrites. It annotates boot, round setup/egg reveal, round bonus tally, intermission, diamond bonus and destroyed-egg/Sno-Bee stun animations. UNKNOWN duration in the original overlay was191.631s; the new exact-tape overlay explains those phases and header/intermission overlaps. This is editorial annotation of the retained native capture, not validation of a generic phase reader or a runtime-agent change.

The egg objective is removed at1376; its later500-point animation1496–1554 and score credit1555 are distinct events. Annotation protocol, phase manifest and video metadata, visual/full-decode checks. Original v1 retained.

The original campaign deadline quoted above belongs to its closed snapshot. Stefano subsequently extended work toward1,000,000, with300,000 as the next score milestone. No new score promotion is made by this review correction.