Project Pengo

Published Pengo milestones

The 500,000 milestone has passed. Before any higher target, a separate first legal 655,360 crossing contract requires word/pixel evidence in BOTH new complete native reset proofs, observed epochs, modular positive awards and small-digit HUD checks. Then 1,000,000 cumulative and ultimately 1,492,230. Legal laboratory rollover remains unobserved. The supplied human benchmark 1,492,220 and our numerical goal 1,492,230 remain distinct.

The supplied human comparison benchmark is 1,492,220, listed for OOO by Twin Galaxies. Its Points rules specify five initial lives; this laboratory uses three. The legacy ROM naming maps to pengoja in recent MAME. Human competition and privileged trajectory search are separate categories. This numerical comparison establishes no human rank or world record.

M03 review Z1–Z3 is applied in a new video v2.1: search-engine provenance and measured phase annotations, with original v1 and unpublished draftv2 preserved. These labels describe this tape only; frozen planner/death/restore readers are unchanged. Review annotation scope.

Latest published milestone: 500,350 verified / M05

500,350 alive / 500,350 at tape end; 4 lives remain, zero detected deaths. 86 chronological rounds completed; ACT 7 of cycle 6 is unfinished with 2 objectives.

TRAJECTORY SEARCH / PRIVILEGED / REAL TIME REPLAY VERIFICATION. 0 MODEL CALLS. Validated healthy-PLAY restore search. Two NEW whole native reset proofs and result-bound pixel/chronological-round gates pass, 0 LOAD / 0 RAM writes. A third whole native reset supplies the complete video and agrees with BOTH proofs.

M05 report and evidence · Complete native replay. Next: a separate preregistered first legal 655,360 crossing contract before targets above 500k. Then 1,000,000 cumulative and the numerical goal 1,492,230.

Historical 300k milestone snapshot: 301,030 verified / M04

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.

Category: TRAJECTORY SEARCH / PRIVILEGED / REAL TIME REPLAY VERIFICATION. 0 MODEL CALLS. Search uses validated healthy-PLAY restore. Two NEW complete native reset proofs, 0 LOAD / 0 RAM writes, pass the independent result-bound pixel and chronological-round gates. A third NEW complete native reset supplies this whole video and agrees with both proofs.

M04 report and evidence · Complete native replay. A10 passes its frozen 300,000 goal. M05 now targets 500,000 with A10’s configuration unchanged. Independent persistent-process fidelity remains separate and is not adopted in the active run.

Historical 200k milestone snapshot: 216,480 verified / M03B

216,480 alive / 216,480 at tape end; four lives remain, zero detected deaths. 36 chronological rounds completed; ACT 5 of the third cycle remains unfinished with one objective.

Category: TRAJECTORY SEARCH / PRIVILEGED / REAL TIME REPLAY VERIFICATION. Search uses validated healthy-PLAY restore. Verification uses two NEW complete native reset processes, 0 LOAD / 0 RAM writes, plus an independent score-from-pixels gate. A third NEW complete native reset supplies the video and agrees with both proofs. This is a verified trajectory prefix, not a completed full game or an autonomous agent.

200k report and verification evidence · Complete native replay. A08 retains its original 300,000 target and NUL outcome for stagnation; publishing the verified 200k prefix does not revise the experiment. A09 continues separately. Next milestone: 300,000.

Historical 100k milestone snapshot: 100,400 verified

Published verified planner milestone: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.

100k report, input tape and verification evidence. The next separately frozen numeric target is300,000. Historical reports and previous milestone snapshots are preserved below.

Historical30k milestone snapshot:31,040 verified

Historical best verified planner trajectory:31,040 alive /31,040 at tape end. Four lives remain and no death is detected in the retained13,149-frame tape. ACT1–4 are complete; ACT5 is unfinished with8 objectives. Two new native-speed fresh reset processes agree on exact input/full declared trace and sampled pixels/devices,0LOAD and0RAM writes. TRAJECTORY SEARCH / PRIVILEGED / REAL TIME REPLAY VERIFICATION, not a full-game/world record or an autonomous agent.

30k report, input tape and verification evidence. The previous search attempt remains NUL: its final ACT4 Sno-Bee entered an unknown boundary state. A separate native diagnostic validates departure, then a19-second GAME TIME payout of5,000points and the30k extra life. A scoped post-hoc reader passes104retained traces before independent score promotion. The next frozen configuration targets100k; intermediate handoffs use accelerated new resets, while FINAL promotion always uses two new native full resets. Models/student/PPO/new games remain suspended under the unique300k objective.

20k and30k are verified;100k and300k are pending. Comparable leaderboard provenance remains unverified. Previous milestone snapshots and frozen reports are retained below.

Historical20k milestone snapshot:21,210 verified

Historical best verified planner trajectory:21,210 alive /21,210 at tape end. Two new MAME processes replay the complete9,184-frame tape from reset at native speed, with no LOAD and no RAM writes. Three lives remain; no death is detected in the retained tape. ACT1 and ACT2 are complete; ACT3 remains unfinished with3 objectives. TRAJECTORY SEARCH / PRIVILEGED / REAL TIME REPLAY VERIFICATION. This is not a full-game/world record or an autonomous-agent result.

Milestone report, input tape and verification evidence. An interior three-diamond alignment in ACT3 produces a measured10,000-point bonus, paid as100 increments of100. Production restore is admitted only for the validated healthy PLAY scope; final verification always starts from reset. Readers reconcile per-act eggs/objects across all16 diagnostic acts; Rack Test remains diagnostic only and supplies no scoring seed.

Stefano’s current numeric roadmap is20,000 →30,000 →100,000 →300,000.20k is reached; a separately frozen attempt now targets30k. A comparable leaderboard remains unverified. Student/autonomous-agent/model comparisons and new games are suspended. Historical results and Pleiads artifacts are preserved below; their statements describe their dated configurations rather than today’s campaign.

Historical research snapshots through6October2026

Final fixed latency sample

The preregistered extension is complete. Within the first30 distinct contacts in fixed recipe order, 23 timelines are complete and attributable and7 are censored by player death. The minimum20-event latency gate passes.37 matching contacts are retained overall, including7 beyond the declared report sample. Timing variants share roots; these are dependent observations, not37 independent starts.

Two native reset replicas of each recipe agree on declared trace bytes, sampled pixels and exposed devices. The combined65 recipes use130 processes and647,650 physical frames. The32-recipe extension costs316,944 frames and658.015seconds. Contact-to-counter delay varies97–121frames;117 is not constant.

All37 derived event-clock records and the declared30-event sample · Retained extension report · Preregistered timing matrix.

This closes the latency prerequisite only. E0056to0 still needs the opt-in controller, validated causal event reader and final preregistration. E017 is running six fixed movement pulses at the skipped-cell state; no new score search or controller installation is claimed. Complete-machine identity remains UNKNOWN. The first-batch video below is explicitly its earlier17-contact/14-usable snapshot; this extension leaves that editorial artifact unchanged.

Pengo is the first implementation of the generic GameAdapter contract. The lab separates privileged planning from a pixels-only symbolic board/ASCII interface. The logical 13×15 grid was measured experimentally before freezing it.

Historical best verified planner trajectory: 9,790 at tape end; 9,390 alive. ACT 1 and ACT 2 complete; ACT 3 incomplete. A death begins at frame 7,299 before the lives counter updates. This is not a full-game record or an autonomous-agent result.

Read the ACT 3 diagnosis and evidence.

The next progression milestone is Three Acts, Zero Deaths: three completed ACTs, three lives remaining and no detected death. Diamond mechanics are a separate later experiment; scoring optimization waits for verified mechanics. 304,000 is a numeric target with unverified comparable-leaderboard provenance.

Research uses local MAME 0.289, pengoja, frozen DIP/difficulty and Python CPU workers on doppia-3090. Historical Project Pleiads artifacts remain unchanged. The pixel controller is a prototype; no Qwen benefit or general robustness has been established.

E003 outcome update

The single preregistered Three Acts, Zero Deaths attempt is incomplete, stopped at progress-timeout with 6 ACT 3 objectives remaining. The retained recorded prefix has a lives counter of 3 and no observed death in an offline audit; ACT 3 completion is not verified. No score search or retry follows. Read the outcome and exact diagnostic captures.

E004 diagnosis update

Hypothesis pending the reader-only counterfactual: the recorded ACT 3 stagnation may reflect incomplete egg representation, planner limitations, or both. The retained segment has no detected deaths and25decisions without official objective reduction. The old reader made all egg targets UNKNOWN; this observation does not isolate the causal bottleneck. The causal claim remains untested in this recorded segment. E004 Objective Reduction Analysis identifies missing ACT 3 egg coverage: the inherited reader scans six entries, while the observed state uses entries 7 and 8 too. Three Sno-Bee and three eggs reconcile the endpoint count. One existing 6→5 ice-crush event is reproduced twice at native speed, with 117 frames from contact to count decrement. Read the diagnosis, geometry and new video. No new search follows. Diamond verification remains separate.

Editorial video review v2

The W1–W6 review is closed: fixed ACT 3: death at frame 7,299, 9,390 alive / 9,790 at tape end, fixed 13,211 TOTAL evaluations, EXPLORATORY PILOT title and explicit trace-hash region coverage. E009 v1/v1.1 are superseded and not publishable; E012 editorial v1 is unlisted and superseded. Original files remain unchanged. Read the review closure and revised video. No new search follows.

Historical E002 review response

E002 remains one ACT, one canonical start, offline trajectory search, 3,850 verified points. The review corrects total work to 4,868 evaluations / 7,219,315 FE / at least 622.203 recorded seconds, with missing times left UNKNOWN. The 81 decision logs include the final rejection. Fidelity stops at applicability F2; N2 has only one distinct maze across 64 START offsets; N1 remains incomplete under its original watchdog. Read all A/Q/V/F/N/R answers or watch the preserved corrected v3.1. No new search follows.

Progress diagnostics6October2026

E005 is now Objective Reduction Planner, targeting a healthy ACT3 prefix6 to0 without optimizing score or time. The earlier unexecuted scoring assignment is superseded and preserved as history.9,790 remains best verified planner trajectory,9,090 the incomplete E003 observation; no new full-game record or completed ACT3 is claimed.

The first E015 native latency batch shows97–121physical frames across14 complete attributable events after strict deduplication.117 is not constant. A separately frozen32-recipe timing extension is in progress; the20-event gate is not yet passed in the published first batch. E016 matched egg controls distinguish hatching, which preserves objectives, from two direct ACT3 removals with immediate official decrements. Later death during one neutral wait remains visible. E014 reproduces a skipped cell target; E013 retains the earlier failed routes.

No scoring search or N1b is launched. Egg display/score timing, residual cell control and event acknowledgement remain separate validation work. Diamonds are prerequisites for future scoring, not a presumed dependency of this non-scoring E005.

Review correction6October2026

The causal interpretation above is a hypothesis. The next controlled trial is E020, E003 with only the egg reader changed, preserving the original controller, heuristic, seed and budgets. The redesigned E005 is deferred. No new scoring or autonomous-agent result is claimed. Future stagnation is NUL for its registered configuration; infrastructure failures are separately incomplete. Frozen historical results retain their original labels.

Y1–Y7 response and exact mismatch explanation. The11E003 and7E009 original arithmetic discrepancies are explained by egg transport and one retirement/hatch boundary. This compares RAM-derived counts with a counter in the same RAM; it is not independent pixel-reader accuracy. Per-act initial counts measured so far are6,6,8 forACT1,2,3; ACT4+ remains UNKNOWN despite the finite Rack diagnostic. The new reader preserves two additional E009 header-update UNKNOWNs.

Historical9,790 editorial v2 already applies W1–W6; the originals remain unchanged and superseded.9,390last alive /9,790tape end, death7299, ACT3incomplete.

Reader-only counterfactual preregistered

E020 frozen protocol and manifest. Same E003 planner, original controller and prefix; only egg observation changes. E020 ended NUL at9,490observed points,3lives and5remaining objectives; the reader fix alone was insufficient in this configuration. No new verified record.

New objective-reduction strategy

E005 frozen protocol and source hashes: causal crush/egg credit, reachable push stances and the admitted opt-in cell controller. One new attempt after the terminal reader-only comparison. Gameplay outcome pending; privileged trajectory search.