← research
CLOSED Seed Model · v0.3

Instinct, Orientation and Acquired Memory

A recorded-data demonstration of a closed development stage. Closed October 10, 2026.

The organism

Instinct supplies local survival and recovery. Orientation preserves an ideal internal reference to the starting pose. Experience can store a bounded food-place anchor; installed return and local-search rules act on it. Location is acquired. The return/search rule is installed.

Installed mechanismssurvival ROM · recovery · orientation · return / local search
Acquired statecandidate-local associations · one replaceable food-place anchor
Later experimental additionsguidance and relative-action profiles — not part of the reference organism

Matched recorded trajectories

EXP-005 replays three matched arms at one transition index. Terminal panels remain terminal; no movement is inferred after exhaustion.

Condition
Illustrative case

Post-hoc illustrative selection: loading paired spatial-minus-ROM reserve difference —. The full campaign, not this example, supports aggregate findings.

Loading recorded arm…
Loading recorded arm…
Loading recorded arm…
transition 0 / 0

World, prey positions and identities are observer information, not controller inputs. The display replays saved transition records only; it does not simulate or extrapolate.

What the experiments established

Restricted reserve is average bodily energy across the fixed 1,024-transition horizon, with zero contribution after terminal exhaustion.

EXP-005 primarylocal-plus-spatial package minus ROM
+0.23854 [+0.15559, +0.31801]
EXP-005 secondaryspatial package added to local association
+0.26874 [+0.19794, +0.33913]
EXP-006 primaryrelative adaptive minus relative frozen
-0.009091 [-0.067395, +0.05298]
EXP-006 secondaryguided minus legacy
-0.145416 [-0.2333, -0.056783]
EXP-007 supplementarydelayed minus short
-0.049811 [-0.110324, +0.008844]

EXP-005 primary is a local-associative-plus-spatial package versus ROM; its secondary more directly tests adding the spatial package. The package does not separate acquired storage from installed steering. Relative learning did not demonstrate its intended reserve advantage; the tested guidance implementation was unfavorable in this condition. A capture is not evidence of learned hunting.

Progression and limits

From static food to moving prey, EXP-005 spatial survivors were 33/64 static and 38/64 moving. EXP-006 legacy survivors were 38/64 static and 33/64 moving; it did not repeat the spatial ablation. EXP-001–004 contain 1,088 evaluation lives; central EXP-005/006 evidence 896; supplementary EXP-007 384; all cohorts 2,368. These cohorts are not pooled into a single efficacy estimate.

What remains open: food/non-food discrimination, sustained search and re-encounter, storage versus steering, guidance utility, and biological benchmark alignment. EXP-007 remains supplementary: its archived sources reproduce the recorded input fingerprint but do not independently establish a complete pre-execution freeze or exact executable.

Evidence and provenance

Reference organism: EXP-005 recovery_local_spatial; historical executable e090753800d40ac90189357daa967de6edc546fe. Source repository links may require access.

Download closure textDownload numerical summaryDownload selected replay indexDownload export manifest