# PENGO-E004 — Objective Reduction Analysis

**E003 failed because the planner could survive the segment but could not
reduce the six remaining ACT 3 objectives. The bottleneck shifted from
survival to task progress.** This describes the recorded segment and frozen
policy, not all Pengo situations.9790 remains best verified planner trajectory;
9090 remains an observation in incomplete E003. No new search was started.

The diagnosis found a concrete representation defect before proposing a new
objective: the historical egg reader scans six coordinate entries. ACT3 has
valid observed entries at indices6 and7 as well. It therefore returns UNKNOWN
for eggs on all1290 E003 ACT3 observations and1282 pre-death E009 observations.
The inherited policy then falls back to visible enemies, omitting known egg
targets. This supports a representation bottleneck; it does not prove that
fixing the list alone would complete ACT3. The planner has not been restarted.

The offline eight-entry decoder is explicitly scoped to observed ACT3. No
later-ACT table capacity is claimed, and no RAM-derived state enters the pixel
agent. At the frozen E003 endpoint7306, player(9,7), the six objectives reconcile
as three mobile Sno-Bee and three unhatched eggs:

| Objective | Cell | Observed motion in preceding64 frames | Nearby pushable ice | Known-empty walk to its stance | Current attack geometry |
| --- | --- | --- | --- | --- | --- |
| Enemy slot1, native raw2 | (5,8) | moved; net(+21,-42) pixels | (5,9), LEFT, stance(6,9) |5 cells | No reachable current one-push lane to target cell |
| Enemy slot2, native raw2 | (5,7) | moved; net(+2,-16) pixels | (5,5), UP, stance(5,6) |5 cells | No reachable current one-push lane to target cell |
| Enemy slot3, native raw2 | (7,8) | moved; net(+14,-41) pixels | (7,6), LEFT, stance(8,6) |2 cells | No reachable current one-push lane to target cell |
| Egg entry3 | (3,13) | stationary in recorded window | its own ice; LEFT from(4,13) |11 cells | Slide stops before ice at(1,13); reduction not verified for this configuration |
| Egg entry4 | (3,11) | stationary in recorded window | its own ice; LEFT from(4,11) |9 cells | Slide stops before diamond at(1,11); reduction not verified for this configuration |
| Egg entry7 | (5,1) | stationary in recorded window | blocked LEFT interaction from(6,1) |11 cells | Ice at(4,1) prevents sliding; candidate break, not a verified elimination |

Coordinates are zero-based logical cells. Nearby ice is not necessarily an
attack on that target. Enemy trajectories, lethal contact and timing can
invalidate static reachability. The full geometry JSON records every aligned
candidate lane, blocking cell, optimistic route obstacle count and uncertainty.
The three eggs have matching native ice tiles at the census frame.

The eight-entry list plus populated native actor slots matches the counter
on1279/1290 E003 and1275/1282 E009 frames.11 and7 transient observations remain
unreconciled; the native header can also lag coordinate changes. These are
explicit limitations, not silently repaired counts. At the26 chosen
stagnation endpoints all six objectives reconcile. The analysis-v1 receipt is
retained with its correction: it wrongly turned UNKNOWN eggs into zero; v2
supersedes that interpretation.

## One verified6→5 event

An existing E009 tape supplies a concrete, ordinary ice-crush example:

| Frame/interval | Observed event |
| --- | --- |
|6771..6772 input selections | RIGHT+BUTTON from player(4,7), ice at(5,7); source lane(6,7),(7,7), obstruction ice(8,7) |
|6773 | Source ice cell becomes empty |
|6778 | Enemy at(7,7) enters native hit state10 |
|6783 | Crush animation state11 |
|6807 | Points state12 and +400 observed; counter still6 |
|6895 | Counter6→5; the retired slot begins replacement hatch at(3,11), removing that reserve coordinate entry |

The count decrement occurs117 physical frames after contact, about1.93s.
Two fresh native-speed reset replays of the unchanged existing tape, ending
7023, match every declared RAM/sprite/driver trace byte and sampled native
pixels/devices. Wall times116.219s and116.235s;14046 physical frames, separate
diagnostic budget. One unique event with two replicas is not two independent
mechanics cases. Complete machine-state identity remains UNKNOWN. No death or
life fault occurs in the tested ACT3 segment. This does not complete ACT3 or
verify any diamond mechanic/value.

## Existing branches and distance proxy

A read-only scan of1639 already simulated endpoints in E003 decisions40..65
finds no endpoint below6, accepted or rejected.38 endpoints at7 belong to
decision40, whose root still had7 before the chosen transition to6; they are
not evidence of a counter increase.1448 accepted endpoints end at6. No new
branch, action, simulation or parameter tuning was performed.

The offline proxy is shortest known-empty walk to a static interaction
stance, with an optimistic lower bound on intervening ice cells. Across the
26 chosen endpoints the best stance distance fluctuates between2 and10 and
ends at9, while the counter remains6. A close stance can concern moving
enemies or an egg slide that does not eliminate it. Therefore this proxy is
not yet true distance-to-next-progress-event, a proven minimum push count or
a sufficient offensive objective. Per-target commitment, block/egg motion,
the native hit→points→decrement delay and explicit incomplete geometry need
to be represented before a new frozen planner trial.

The concrete next engineering prerequisite is measured ACT3 egg coverage;
then a separately preregistered policy can evaluate event-oriented progress.
No such policy is installed by E004. Diamond mechanics stay separate and
unverified; E005 scoring and304000 optimization have not resumed. The earlier
unexecuted E004 diamond plan is superseded in experiment purpose by the user's
post-E003 instruction. Original N1's watchdog still takes precedence; N1b is
not adopted. Historical Pleiads/Pengo sources, results and tapes are unchanged.

Evidence: data/pengo/e004/objective-analysis-v1 and v2, confirmation-v1,
event-details-v1, existing-branch-scan-v1 and counter-anomalies-v1.