The state turns before the form.
Without seeing the eventual transition, we interrogated each Flow Lenia state with eight exact inventory-preserving rearrangements. Whole-over-parts organization rose. Φ and ordinary predictability stopped wanting the same future. The turn usually arrived before the organism became visibly itself.
The two objectives separate.
At step 16, Φ-directed rearrangements still tended to improve TDMI too. By steps 40–48, the same objectives were neutral or opposed. These checkpoints were fixed before any new outcome existed.
Whole-over-parts rises
Mean among organisms whose transition is still ahead.
Φ / TDMI alignment rotates
Spearman alignment across eight local state probes.
The turn happens inside individual lives.
Among the same 48 organisms still pre-transition at step 48, whole-over-parts rose while action alignment and the TDMI response along the Φ direction fell.
Ethereal cell 1549 · transition at step 63
Serene dancer 0331 · transition at step 73
A coordinate, not an exact countdown.
The information-geometry model preserved rank information and often moved monotonically, but its exact timestamp error did not beat morphology and clock controls.
Whole-over-parts versus actual progress: ρ +0.523. Alignment versus progress: ρ -0.414. First negative Φ-directed TDMI response appeared in 51 organisms.
Controls-only MAE: 39.88 steps. Controls plus information geometry: 40.46. The added coordinate describes phase better than exact time-to-transition.
Intervene on the turn itself.
For each fresh organism, wait until the Φ-directed TDMI response first becomes negative. Intervene then. Compare with the exact same action delivered at the same absolute time in organisms whose geometry has not yet turned. If transition timing or adult form changes by inferred phase rather than clock time, we have isolated a causal developmental state.