At the structural peak, a single Φ-directed perturbation made eight matched unseen futures converge on fine morphology. It did not simply pick a bigger body, preserve a founder, or reduce fragmentation. It narrowed what shapes the same organism could still become.
−4.55%
pairwise shape distance, Φ-forward vs Φ-backward
13 / 19
family means narrowed in the same direction
P6 → P12
continuous late-passage separation
01 · The direct swing
Touch the same organism at three developmental ages.
Every organism had its own previously measured structural peak. We intervened well before it, near the turn into it, or exactly at it. At each age, sixteen equal 20% inventory-preserving swaps were scored from the state’s past alone. Then four policies were released into eight fresh, matched futures.
same living state
38 organisms19 organic families × 2 schedules
→
one local physical push
Φ-forwardΦ-backwardblind swapuntouched
same dose · different information direction
→
future fan
8 unseen worldsshared randomness · 12 later passages
Peak − 8early intervention · branch at peak − 6
Peak − 4turn intervention · branch at peak − 2
Peaklate intervention · branch at peak + 2
The selector did not ask which swap looked healthy or creature-like. Φ-forward chose the swap that moved the local state furthest toward higher whole-over-parts prediction while moving ordinary whole-system predictability the other way. Φ-backward chose the opposite end of that same local possibility set.
02 · What did not resolve
The main body was not the timing gate.
The frozen primary asked whether Φ-forward would make unseen futures agree about how much mass consolidates into one dominant component. The direction was modestly positive at all three ages, but every 95% family-bootstrap interval crossed zero. There was no clean early, turn, or late aperture on this coarse body-allocation measure.
Canalization of dominant-body allocation · backward minus forward future breadth
That matters. The state was not giving us a generic “more Φ means a more settled organism” effect. If information geometry was steering anything here, it was steering a more specific layer of development.
03 · The discovery
At the peak, Φ-forward narrowed fine morphology.
Fine morphology was a predeclared secondary panel. We compared the eight future density fields after translation alignment, then asked how many effectively distinct shapes remained. Lower values mean the futures resemble one another more strongly.
Effective shape repertoire · passages 5–12 · lower means more convergent
Φ-forwardblinduntouchedΦ-backward
Late mean pairwise shape-distance contrasts
Futures became more alike
0.08364 vs 0.08763
Mean pairwise shape distance, Φ-forward versus Φ-backward. Difference −0.00399; 95% interval [−0.00713, −0.00117].
Not just the opposite arm
−0.00340
Φ-forward versus untouched pairwise shape distance; 95% interval [−0.00659, −0.00023]. Φ-backward and untouched did not separate.
Illustrative only: f45 / schedule 129, chosen mechanically as the negative late cell closest to the full cohort mean—not as the most dramatic example. F1–F8 use the same eight future seed pairs across rows. The difference is subtle by eye because the effect is a cohort-level compression of pairwise morphology, not a cartoonishly different phenotype.
Late comparison
Shape-repertoire change
95% interval
Family direction
Φ-forward − Φ-backward
−0.004257
[−0.007554, −0.001281]
13/19 narrower
Φ-forward − untouched
−0.003607
[−0.006887, −0.000441]
13/19 narrower
Φ-forward − blind
−0.001851
[−0.004167, +0.000289]
12/19 narrower
Φ-backward − untouched
+0.000650
[−0.002598, +0.003777]
direction unresolved
04 · It develops
The divergence is delayed, then keeps growing.
The future fields do not split immediately after the intervention. The forward–backward difference builds gradually. From future passage 6 through passage 12, every fixed passage interval is above zero in the canalization direction.
Late morphology canalization over future time · backward minus forward effective repertoire
This does not look like an immediate geometric scar. The physical swap happens once. The convergence becomes legible several organism passages later and strengthens as the futures unfold.
Family means · late Φ-forward minus Φ-backward pairwise shape distance
05 · What kind of closure?
A shape constraint—not identity collapse.
The same late contrast did not resolve on founder composition, dominant-body fraction, dominant-body breadth, or component-count breadth. The anchored and translation-normalized morphology views were essentially identical, so ordinary drift across the wall is not the explanation.
Shape converged
Translation-normalized repertoire −0.004257 and wall-anchored repertoire −0.004256, both with intervals fully below zero.
Founder mixture did not
Founder-composition repertoire +0.000115, interval [−0.002725, +0.002769]. The lineages were not simply homogenized.
Coarse body organization did not
Largest-body breadth −0.00287, largest-body mean −0.00426, and component-count breadth −0.385 all remained unresolved.
What is causally supported
Given the same starting state, same perturbation dose, and same unseen future seeds, selecting the local Φ-forward direction at the structural peak compressed the later ensemble of fine morphologies.
What is not yet settled
The late arm is the one with a clear within-age effect, but late-minus-early and late-minus-turn intervals still cross zero. We have a late-phase causal morphology signal—not yet a proven timing interaction.
06 · The larger picture
Information geometry is acting like a constraint on possibility.
This is closer to Levin’s motivating idea than a generic predictor would be. A property measured from the organization of the present state identified a physical direction in state space. Moving the organism along that direction changed the multiplicity of later forms available under environmental variation.
Φ did not forecast one ordained creature. It marked a direction that changed how many morphological answers the future could give.
Not merely correlation
The selector chose among sixteen real, equal-dose perturbations. Forward and backward were then run under matched future randomness.
Not generic stabilization
The effect was selective to fine morphology. Identity mixture and coarse body organization remained open.
Not a trivial estimator artifact
The one zero-leverage state produced exactly identical forward and backward futures: maximum scientific difference 0.0.
Stop here and think
The original timing hypothesis was too coarse. The experiment uncovered a more interesting possibility: developmental commitment may be multidimensional. “Which body wins,” “which founders persist,” and “which exact shapes remain reachable” can close at different times and respond to different information directions. No next experiment has been launched.
07 · Scale and boundaries
Large enough to see structure. Specific enough to stay honest.
3,648complete future trajectories
43,776future passage states
19 × 2families × schedules
20,000family-bootstrap draws per interval
The dominant-body timing question and all morphology, founder-composition, and component-count panels were fixed before outcomes opened. The morphology signal is a predeclared secondary result. Additional timing contrasts and the representative fan are source-bound post-outcome descriptions. This is a causal intervention result in this Flow Lenia cohort—not autonomous replication, a universal causal-emergence theorem, or proof that Φ is uniquely sufficient.
