Pengo E005: causal objective reduction planner

Preregistered; gameplay result pending. Frozen at2026-10-06T20:05:07.051499+00:00, full private manifest SHA256 bf85cf27665f630584bb5dbbb9918631a2acb9a00ab3557f929fd746af4cbbb6. This new configuration seeks ACT3 objectives6→0 with3lives and no detected death. Score is observed, never ranked. TRAJECTORY SEARCH / PRIVILEGED; no autonomous-agent result is claimed.

The root is the original E003 reset tape clipped at7049: healthy retained prefix, aligned player, all195board cells known,6official objectives. The previously proposed6876root contains two UNKNOWN cells and is rejected. Two fresh native-speed root replays must match the admitted observed prefix before planning. No Rack-derived prefix is used.

New strategy: reach a stance behind usable ice, verify a clear push lane with a compatible current target, or approach a blocked egg block for direct destruction. Empty-cell reachability measures distance to an attack setup. Six current setups are visible at the root. Future enemy crossing is uncertain, so geometry alone is not a kill claim: exact forward simulations must confirm an event.

The prefix-only causal reader logs contact, confirms a crush in the known animation/context and gives one unique credit. The later official counter decrement is acknowledgement, with no second credit. Direct egg destruction requires observed applied push, correct stance, blocked ice and matching entry/counter removal. Slide and hatch alone are not credited. The retained admission covers30native crush events (23complete,7death-censored),10native egg/control/hatch traces, E003/E009 transport reconciliation and two verified direct-removal stance geometries. Unsupported simultaneous or unexplained cases remain UNKNOWN.

The new opt-in E018v2 cell controller has nine admitted native cases and a measured legacy direction-entry exception. These gates are scoped examples, not proof of every geometry. Historical controllers, frozen results and121source contracts remain unchanged.

One attempt:20workers,width6/depth6,80million physical frame evaluations,600decisions,3600search seconds and40committed decisions without a new causal credit. Stop on the first healthy counter0 candidate, then a fixed1800neutral-frame transition probe and fresh native reset verification; no ACT4planning. Stagnation and ordinary budget stops are NUL, actual infrastructure failure INCOMPLETE. A null configuration does not cancel the user’s overnight research; a later strategy needs a new preregistration.

Frozen executable protocol · Source-hash manifest. An initial preparation failed while hashing an E004summary that had not been copied to the remote machine:0MAME and0search frames. The original empty preparation namespace and failure receipt remain preserved; v1.1 changes only preparation namespace/source filename, with identical game policy and bounds.

Rack timing diagnostic v3 ran separately after E020, before this attempt: six fresh replicas,180000frames, matching declared observations. The pulse reproduced ACT1=6,ACT2=6,ACT3=8initial capacities. Hold variants admitted no later acts; ACT4+ remains UNKNOWN. Rack is DIAGNOSTIC ONLY, never a score or gameplay result. Diamond mechanics/value remain unverified and are excluded.304,000 remains a numerical target with no verified comparable leaderboard.

Current best verified planner trajectory remains9,790 at tape end;9,390last alive and death from7299. Corrected exploratory pilot video. Any improved candidate will be reported only after fresh replay verification.