Completed development · native GAM · descriptive

What varied
with experience.

These surfaces are stored predictions from models fitted after one completed developmental lineage: 491,520 physical transitions from four episodes. Three worlds in the first three episodes trained the models, a fourth world validated them, and the entire fourth episode was held out.

They describe conditional structure in this run. They do not establish causes, welfare, intent, or a capability in a new life.

Effort · whole episode held out0.0201RMSE · mean baseline 0.1323
Body-law residual · whole episode held out0.3430RMSE · mean baseline 0.3451
Goal progressNo model mintedNative optimizer did not certify a stationary fit.

Actual fitted surfaces

Hold one life still.
Move two coordinates.

Each 25 × 25 surface varies the named pair while other features remain at their recorded means. Hover a cell for its native axis values and prediction. The display rescales color within each surface; the numeric tooltip retains the artifact's units.

Loading compact atlas…

Behavior audit · all transitions

Contact happened. Regulation did not.

90 of 96 resident-episode trajectories had mouth-material contact, totaling 1,853 contact steps and 28.35 units of ingested mass.

Every episode remained energy-negative. All 96 resident-episode physical reward sums were negative. The learned controller acquired material, but did not sustain its bodies over the 256-second lifecycle.

Read the full behavior audit ↗

Rows: 276,480 train · 92,160 validation worlds · 122,880 whole-episode held out

Source: full native artifact ↗ · compact public surface data retains the source artifact SHA-256