Pengo30k milestone:31,040 verified

30,000-point milestone PASS:31,040 alive and31,040 at tape end,4lives, no detected death in the retained13,149-frame tape. ACT1–4complete; ACT5unfinished with8objectives. Category: TRAJECTORY SEARCH / PRIVILEGED / REAL TIME REPLAY VERIFICATION. This is best verified planner trajectory, not a full-game/world record or an autonomous pixel agent. The ROM and all frozen PleiadsE007/E008 artifacts remain untouched.

A null attempt led to a reader diagnosis

M02attempt01 remains NUL, stopped by its40-decision progress watchdog at26,040observed points:44plans,78commits,3,378search nodes,3,382MAME launches and115,070physical frames in772.282seconds. It had completed ACT3 and reached ACT4with one objective. Late branches were rejected because the final Sno-Bee entered raw state17, missing from the reader. This was not evidence that the planner could not eliminate the last enemy. We stopped scoring and ran E040, a separately frozen microscope from the same healthy witnessed leaf11,596.

E040 compares neutral input and all five legal macros PU/MD/ML/MR/W0, each extended by1,800neutral frames and independently replayed twice from reset at native speed. All12replicas agree on actual input, complete known prefix, sampled pixels/devices and their paired traces,0LOAD/RAM writes. Five diagnostic macro transports plus12native jobs cost161,045physical frames/17MAME launches and279.968seconds. First UNKNOWN screenshot, tile/object state and before/after RAM diff remain private; no raw RAM/state dump is distributed.

The neutral pair shows state17at11,601, departure sprite codes49→55 at the bottom-left boundary, and objective1→0at11,721. The actor is gone; no block contact or score increment occurs at removal. PU reveals a one-frame18retirement at11,724 before0. Delay from17entry to official removal is120–123frames across the observed variants, not a universal constant. Pixel departure and actor/counter history support BOUNDARY_ESCAPE, distinct from crush/kill credit.

Five thousand points and a fourth life

The neutral continuation pays5,000points at11,913, changing score26,040→31,040. The rendered GAME TIME screen says19seconds and displays the0–19second5,000point bracket. This is the observed ACT4time payout; no unmeasured cross-act numeric reward potential was inserted into the search. The30k threshold also changes the life byte3→132(0x84) and HUD reserves2→3:4total lives. The unchanged old reader treated132as UNKNOWN.

E041 is explicitly post-hoc scoped reader admission, then frozen before regression. It recognises only the witnessed ACT4bottom-left final-actor17/18context;17remains an objective until official retirement. Its escape removal is separate from ice/diamond crushes and gives no invented game score. Only raw0x84with valid HUD score≥30k and exactly three reserve glyphs decodes as4lives; other high bytes and mismatching HUD stay UNKNOWN. All104retained traces pass with0added faults, clone consistency and preserved older event/credit/UNKNOWN behaviour. Synthetic interior placement,extra objective,wrong colour,wrong predecessor/no prefix,death latch,corrupt HUD/score/high byte are rejected. Synthetic negatives are annotations, never game results. Reader protocol and sanitized admission. Rack diagnostics stay diagnostic only; no Rack-derived seed is used here.

Independent native promotion

After reader admission, the FIRST healthy aligned ACT5decision point at13,149 is selected retrospectively from E040’s existing neutral trace and frozen before two NEW native fresh reset confirmations. All inputs after11,596 are explicitly neutral and part of this new candidate. The original M02NUL and full E040 diagnostics are preserved; they are not relabelled as successful game runs.

Both new replicas match the exact actual input tape, full declared candidate trace and each other, sampled rendered pixels and exposed CPU/driver registers. Each uses0LOAD and0RAM writes. The complete retained prefix has no detected death; the endpoint is healthy controllable ACT5,4lives,31,040 and8reconciled objectives. Native replicaA takes216.949709seconds; the video runs30.303fps/216.975seconds, including a final odd-frame snapshot. Verification protocol, registration, sanitized proof, complete input tape, video metadata.

Declared trace SHA256 92aa052be601bdff010cc8deb4a0e249fcf2a3d6f6ddefc6bf14ed2ec9ff69aa covers the PENGO01header,physical frame indices,4,096bytes RAM0x8000–0x8fff,16sprite-coordinate bytes and5driver flags. CPU,scheduler,timers,other device state,inputs and pixels are excluded from that hash. Exact inputs,sampled pixels and exposed CPU/driver registers are checked separately. Complete-machine identity remains UNKNOWN. Video SHA256 6b80ad4763f3315b479b3ecd288801cbd281290c7753218afd55b0c700350660. No raw RAM dump,save-state,ROM,model weights or secrets are published.

All work stays counted

The fixed milestone snapshot totals 83,342,992known physical frame evaluations and18,320owned Pengo MAME launches, including counted diagnostics,nulls,preflights and independent final proof. Reused captures and Pleiads work are excluded. Independent30k verification adds26,298physical frames/2processes; E041regression adds0MAME/0frames. These totals are displayed from the video beginning, fixed rather than falsely progressive. Cost snapshot, visual validation.

Baseline remains pengoja/MAME0.289,3initial lives,Medium,Upright,bonus life30k,RackOff,DSW1=176/DSW2=204, native60.606Hz. The fourth life is earned in game. Production restore remains limited to validated healthy initialized PLAY, with final verification from reset. The previous 21,210milestone and its scoped ACT3interior10,000diamond bonus stay historical; wall/cross-act/repeated diamond values remain unverified.

The next separately frozen configuration targets100,000, then300,000. It uses20CPU workers and two accelerated new resets for intermediate ACT handoffs to address the measured minutes per native handoff; transitional states are quarantined and never restored. The next PLAY seed is rebuilt in another fresh process with exact prefix matching. Every selected handoff is covered by the FINAL two new native full-tape confirmations. Intermediate accelerated checks are search evidence, not native gameplay verification. 100k protocol. Normal stalls remain NUL, diagnoses precede new counted configurations. Student,EmbeddingGemma,Qwen,PPO,new games and speed-record experiments remain suspended under the unique300k objective. Comparable leaderboard provenance is unverified.