Search
Twelve deterministic rearrangements at each of three doses: 4%, 12%, and 25% of eligible occupied cells.
A fresh causal exploration · 576 complete futures
Before the organism
We rearranged a developing Flow Lenia seed after step 8 without adding mass, deleting mass, changing its genotype, or changing which cells were occupied. The exact future changed more and more. The organism-level developmental program barely moved.
Event-time identity0.981 ICC
Exact-state divergence at step 9000.359 Hellinger
64fresh organic specimens
9matched futures each
576/576certified organisms formed
518kframes read
01
A clean nudge
At the boundary after step 8, the operator swaps complete matter and parameter vectors between neighboring occupied cells. It conserves total mass, every cell-state vector, the parameter inventory, and occupied support. Only their local arrangement changes.
Choose
Continue the action with the highest Φ-R, lowest Φ-R, highest ordinary TDMI, or a blind hash-selected action. No later morphology was read.
Release
One action, then 891 untouched steps. All arms share byte-identical frames through step 8.
No paper-shaped gate: every arm, every dose, every event, and every morphology trajectory stays in the atlas. The experiment asks what the system does, not whether one favored sentence survives.
02
Three clocks separated
The controller moved the intended readout. That separation then collapsed. But the exact fields did not return: their distance from the untouched world kept growing. Above that microscopic drift, the organism’s developmental timing stayed stable.
Φ-high minus Φ-low
Mean whole-over-parts difference. The action is strong through step 12, fading by roughly step 24.
Microscopic divergence keeps growing
Mean Hellinger distance from each specimen’s untouched field, averaged over all eight intervention arms.
Not one memory, but threeThe information readout forgets quickly. The microstate never forgets. The organism-level program absorbs both.
03
The organism wins
Across nine substantially different exact futures, most variance in birth time and adult form belongs to which organism we started with—not which early rearrangement it received.
Identity fraction of variance
Descriptive intraclass identity across the nine matched futures.
Event timing, medium dose
Each dot is one organism. Φ-high and Φ-low event times cling to the diagonal; orange marks the exceptional 194-step split.
The event-time ICC is 0.981; adult radius is 0.983; adult rectangularity is 0.957. Meanwhile the exact field reaches a mean Hellinger distance of 0.359 from untouched by step 900. This is developmental canalization in a literal dynamical sense: microstates fan out while organism-scale observables remain on a narrow program.
04
Robust is not rigid
Most organisms barely moved in event time. A few sat near a developmental edge. The clearest one took 408 steps under Φ-high, 513 untouched, and 602 under Φ-low—a 194-step high/low span produced by one early inventory-preserving rearrangement.
Population timing response
| dose | mean Φ-high advance | family bootstrap 95% | high earlier / same / later |
|---|---|---|---|
| .04 | +0.94 steps | [-1.62, +4.20] | 29 / 11 / 24 |
| .12 | +3.11 steps | [-1.41, +10.27] | 31 / 15 / 18 |
| .25 | +2.33 steps | [-2.25, +7.61] | 30 / 5 / 29 |
The mean timing shifts are small and uncertain across families. The medium dose is the most suggestive (+3.11 steps; 31 earlier, 15 identical, 18 later), but the giant responder matters. The sharper phenomenon is heterogeneity: a canalized population containing sparse susceptibility points.
05
What Φ did—and did not do
The selector really controlled the early readout. At dose .12, Φ-high minus Φ-low was +0.718, positive in all 16 families. But ordinary predictability moved with it by +2.040, and Φ-high chose exactly the same action as TDMI-high in 35/64 organisms.
Real control
The intervention was not too weak. It created large, dose-ordered immediate information differences and permanently different exact futures.
Brief control
The selected Φ advantage mostly disappeared before morphogenesis. A one-shot early scalar push did not become a self-maintaining information regime.
Coupled geometry
At this early stage, whole-over-parts and whole-state predictability often favor the same local rearrangements. Better partitions or different actions may separate them.
The conceptual moveCausal emergence may be visible less as “a scalar makes birth happen” and more as an organism-level invariant surviving the explosion of its microscopic futures.
06
Where this opens
The atlas changes the question. We now have a concrete organism-level phenomenon to explain: strong macroscopic invariance, persistent microscopic divergence, and rare points where the invariant becomes pliable.
Map the basin
Repeat the same inventory-preserving action at steps 4, 8, 16, and just before each specimen’s predicted transition. Measure where canalization is strongest and where it breaks.
Target susceptibility
Use fresh organisms to test whether early whole-over-parts state predicts the rare large timing responses—before looking at their futures. The question becomes where the developmental program is steerable.
Change the parts
Quadrants may be the wrong causal decomposition. Score body-centered, multiscale, and dynamically learned parts, then ask which decomposition best predicts basin escape versus harmless microscopic drift.
Branch futures
At matched early states, create many tiny futures. Ask whether Φ marks a narrow future repertoire, a wide one, or a boundary between developmental basins.
Test recovery
Levin’s GARD intervention changed persistence more clearly than first-replicator timing. Perturb after the organism forms and ask whether information structure predicts restoration of its developmental identity.
Cross the substrate
If the same micro-divergence / macro-invariance geometry appears in other Lenia rules, cellular automata, and chemical systems, that is a much larger result than a metric tied to one morphology detector.
What happened, plainly: We causally moved an early whole-over-parts information signal with an exceptionally clean perturbation. The scalar movement was brief and did not broadly control organism-birth timing. But the perturbation exposed a stronger phenomenon: exact futures diverged indefinitely while organism timing and adult macro-form remained strikingly invariant, except at sparse susceptibility points. This is not a failure of the causal-emergence search. It is a much more interesting map of where organism-level causation may live.