# PENGO-E004 — Objective Reduction Analysis, protocol v1

User steering after E003 changes E004's purpose. The earlier unexecuted
E004 diamond-mechanics plan remains historical; diamond verification remains
a separate prerequisite before scoring. E003's frozen policy/result is unchanged.

Question: What exact game events reduce the ACT3 objective count, and what
state/action configurations make those events reachable?

This is diagnosis, not new planning or score search. No new candidate actions,
parameter tuning, RAM writes, save/load, model calls or diamond objective.
Only existing recorded reset tapes/observations are analyzed. Any emulator
work is a fixed reproduction of an existing selected event, never a new branch.

Sources: the retained E003 full trace to7306, its frozen journals and tapes;
the E009/E012 neutral diagnostic's pre-death ACT3 prefix6017..7298. Record all
source hashes and the E003 source/protocol registration hashes before analysis.
Enumerate recorded ACT3 counter decreases without assigning their cause in
advance. Specifically inspect the first healthy6→5 in the E009 prefix. For
E003 inspect the26 decision journals40..65: offline endpoint reads of already
paid-for branches, at most their existing files. Report whether accepted
counter-reducing branches existed, rather than assuming none did. This scan
is not a new search. If no eligible event exists, report that result; no new
planning or adaptive input probes are authorized by this protocol.

Census at first6-objective chosen state6876 and endpoint7306, and all chosen
decision frames between them. Reconcile counter6 against active enemy slots,
pending removal/points slots and unhatched egg coordinate entries. Do not
force six distinct visible entities if the evidence cannot account for them.
Slot identity may be reused after hatching/removal; positions alone are not
permanent IDs. Unexplained count differences remain UNKNOWN.

For every accounted objective record position, observed raw/derived state,
empirical mobility in the recorded window, nearest statically pushable ice,
candidate launch stance/lane, blockers, known-empty player path distance,
and observed player/target motion. Distinguish pushable ice from a blocked
break interaction. Diamonds and UNKNOWN cells are not presumed traversable.
Lane intersection is geometric only: moving targets, lethal contact and
timing can invalidate it. No geometric configuration is called sufficient
for elimination before an observed event verifies it.

Record proxies at all26 chosen endpoints: shortest known-empty walk to a
candidate interaction stance; a lower bound on ice cells obstructing an
optimistic route; static attack-lane candidates. These are not proven minimum
push counts or true distance-to-next-progress-event. Correlation with actual
counter drops is measured retrospectively, not assumed. Do not install a
new planner objective under this analysis protocol.

Event windows: up to256 frames before and128 after each selected decrease,
clipped at the first death/unexplained phase or source endpoint. Preserve
input intervals, score changes, egg-table changes, native object transitions,
block trajectory, count timing and life/death audit. No diamond explanation
or novel ACT3 rule is imported from memory.

After offline selection, freeze a separate confirmation receipt with exact
existing tape bytes/hash, event frame, end frame and sample times. Two fresh
native-speed MAME processes replay that fixed reset tape (max240s each).
Compare the declared full RAM/sprite/driver trace and sampled native pixels
with each other and the original prefix; complete machine identity remains
UNKNOWN. Two replicas of one event are one unique event, not two independent
mechanics examples. No extension or changed input follows a failed replay.

Outputs: a six-objective reconciliation (or explicit mismatch), per-object
geometry/motion table, event timeline, transition evidence, conservative
distance proxies and diagnosed limits of the current heuristic. E003 remains
incomplete,9090 only observed;9790 remains best verified planner trajectory.
No budget remainder is spent on automatic planning or retries.