At steps 24 → 48, the turned state rose from 25% to 75% in an entirely new cohort. Thirty-three organisms crossed into it; only one crossed back.
Before the replicator · direct causal swing
The readout turns.
The actuator does not.
A fresh causal experiment reproduced the hidden information-state rotation before visible form—then showed that simply maximizing Φ is not the lever that drives the later transition. The developmental coordinate is real. The control knob is elsewhere.
64new organic specimens
2,048future-blind action probes
832full causal futures
748,800analyzed future frames
01 / RESULT
Two discoveries, not one failed claim.
The Φ-high versus Φ-low timing effect changed sign and remained uncertain across the four fixed clocks: step 24: +3.3 · step 32: +18.4 · step 40: -4.5 · step 48: -2.8.
Two untouched organisms never transitioned. The same intervention family awakened one in 11/12 branches and the other in 0/12.
Φ geometry behaves like an order parameter—a way to read where development is—rather than a one-shot control parameter that reliably pushes development forward.
02 / REPLICATION
The hidden state turns again.
This was not fitted after seeing the futures. “Turned” meant one exact thing: along the local action direction that most increases whole-over-parts information, ordinary TDMI no longer rises with it.
Across the same 64 specimens from step 24 to 48, whole-minus-parts information moved upward by +1.003; Φ/TDMI action alignment moved by -0.561; TDMI-along-Φ moved by -1.985.
This independently repeats the earlier result: the action geometry reorganizes while the coarse creature is still on its way to the detected form change.
03 / CAUSAL TEST
Reading a phase is not the same as steering it.
At each fixed clock we took the exact same organism prefix and branched it four ways: untouched, Φ-high, Φ-low, and a fixed score-blind action. Then we waited to step 900.
If the rotated Φ geometry were itself the developmental lever, Φ-high versus Φ-low should acquire a consistent timing effect after the turn. It did not. All four interval estimates overlap zero, change sign, and are sensitive to rare organisms.
state readout ≠ actuator
developmental coordinate ≠ force along that coordinate
instantaneous Φ increase ≠ guaranteed later transition advance
developmental coordinate ≠ force along that coordinate
instantaneous Φ increase ≠ guaranteed later transition advance
04 / WHAT MOVED
Causal leverage is sparse, nonlinear, and organism-specific.
Most interventions changed the detected transition by only a few steps. A small set of slow or stalled organisms produced enormous advances, delays, or action-dependent splits.
In a descriptive check, current rectangularity predicted the range of transition times across the three actions at every clock (Spearman +.49 to +.68). The signed information direction did not predict the signed Φ-high timing effect (about 0 at every clock).
That tells us where the next science is: not “push Φ harder,” but map the basin’s causal tangent—what perturbation direction actually moves the organism along the developmental manifold.
05 / BASIN CASES
Same clock. Same operator. Radically different futures.
These are not selected success stories; they are the four largest late/stalled causal-leverage cases. Each row begins from an identical prefix. The first two never crossed the detector untouched. Source 35 was inducible; source 16 was inert. Sources 21 and 26 were slow developers whose transition timing split sharply by action.
radiant-orbiter-0626
action at step 33 · state unturned · common view at step 250
serene-dancer-1806
action at step 33 · state unturned · common view at step 250
harmonic-cell-1709
action at step 41 · state turned · common view at step 500
serene-spiral-0762
action at step 41 · state turned · common view at step 500
06 / MODEL
A better picture of the developing organism.
Whole-over-parts organization and Φ/TDMI action geometry rotate before the visible transition. This now appears in another fresh cohort.
The same nominal action can be absorbed, amplified, reverse timing, or wake a stalled trajectory. This susceptibility is sparse and state-dependent.
The detected morphology is the downstream expression. It is neither identical to the information coordinate nor determined by one scalar intervention.
07 / NEXT SWING
Steer the tangent, not the readout.
The clean next experiment is a developmental-tangent controller. On fresh organisms, learn only from short counterfactual probes which action moves the full state toward the empirically replicated phase direction—not merely which action makes Φ largest for four transitions.
Then branch each state with: tangent-forward, tangent-backward, Φ-high, score-blind, and untouched. The target remains independent transition timing and later morphology.
- Freeze the phase direction from the two fresh cohorts: rising whole-minus-parts, falling Φ/TDMI alignment, falling TDMI-along-Φ.
- Choose actions by multivariate movement along that direction over several short steps.
- Stratify, prospectively, the late/stalled “poised versus inert” basin without inventing a success threshold.
- Ask whether tangent-forward actions wake stalled forms, advance ordinary developers, or instead reveal several distinct developmental manifolds.