# PENGO-E003 — Three Acts, Zero Deaths: incomplete

One preregistered attempt ended at `progress-timeout`. ACT1 and
ACT2 are complete through explicit reuse of the frozen reset prefix to6017.
ACT3 remains incomplete: 6 objectives at
frame7306, with a lives counter of3.
The endpoint score9090 is an observation,
not a score objective or a new verified record.

66 committed decisions, 4124 evaluated nodes,
27797990 physical emulator frames including repeated reset
prefixes; 2198.906s measured search wall time.
The final consecutive no-progress count is25.
The fixed limits were50 million physical frames,600 decisions,40 minutes,
25 playing decisions without objective progress. No limits or policy were
changed during the attempt and no retry was launched.

Every applied candidate suffix was screened frame by frame. The retained
recorded prefix has no observed death/lives fault in a separate offline audit.
343 candidate suffixes were rejected by
the phase gate: 156
collision-delay deaths, 147
death animations and 40 unexplained
phases. UNKNOWN is not admitted as safe; no general reader-completeness claim
is made. This does not promote an incomplete trajectory: there is no
ACT3-clear candidate and no two-native-replay E003 milestone verification.
Complete machine-state identity remains UNKNOWN.

The phase reader identifies the collision freeze before the lives decrement.
Historical readers, Pleiads artifacts and original experiment results remain
unchanged. The old9790 trajectory still is **best verified planner trajectory**,
not a full-game record or a zero-death trajectory. Its death starts at7299,
with lives3→2 at7601 in the fixed diagnostic continuation.

N1's25-decision watchdog was already preregistered and N1b is not adopted.
304000 remains a numeric target with no verified comparable leaderboard.
E004's diamond mechanics/values remain unverified; E005 scoring is not started.
No score target, score tie-break, diamond objective, model calls, RAM writes
or state loads were used in E003.

Evidence: data/pengo/e003/three-acts-v1 retains registration, all journals,
chosen tape and original result; closure-v1/result.json is the later offline
summary. Frozen protocol SHA256: 8b2491dde8ae3c5978a699e75bfb153f799da126b2645ae5fab2e3ff94b23a76.
Candidate trace SHA256: cd75a268353874a5e1dc264d8c876a2597bd705516d0a495465b192582db4894.
The input tape and raw observations remain private; selected counts/hashes are public.