Before the replicator / direct swing 0323 Aug 2026
Closed-loop developmental control

The gate stays open.

One strong early intervention—followed mostly by restraint—held Flow Lenia’s causal-developmental degrees of freedom open for eight passages after release.

39organisms
6,240candidate futures
4feedback decisions
8passages of memory
01 / The controller discovered a strategy

Hit early. Then get out of the way.

At every passage, the controller tried five spatial pulse sizes and asked which future preserved the greatest separation between alternative causal histories. It was free to keep forcing. It usually chose not to.

Mean chosen dose by passage

0255075100P1P2P3P4
openclose

The asymmetric recipe

Open: 25 of 39 organisms chose the full pulse at passage 1. Then 28, 32, and 28 chose zero at passages 2–4.

Close: only one chose full strength at passage 1. Later, heavy doses dominated.

119mean total open dose
225mean total close dose
1.9×more dose to close
02 / Release test

The controlled difference survived after control ended.

Passage 4 is the score the controller directly optimized. Passages 6, 8, and 12 were untouched futures. The open and closed worlds remained apart throughout.

Open minus closed: causal-history separation

0.000.010.020.030.04P4P6P8P12CONTROL ENDS ↑

Eight passages later

+0.018still separated at passage 12

95% interval [0.013, 0.023]. The direction was positive in 34 of 39 organisms.

The cumulative post-release difference across passages 6, 8, and 12 was +0.075, interval [0.059, 0.090].

Every one of the 16 organism families showed positive accumulated post-release separation.
16/16independent family directions
03 / Feedback mattered

Blindly hammering the system did the opposite.

The full-strength pulse applied at every passage did not hold the gate open. It pushed causal-history separation below the untouched world. The adaptive controller beat blind full forcing at every measured age—including eight passages after release.

+0.036open over fixed-full at passage 4
+0.034open over fixed-full at passage 8
+0.026open over fixed-full at passage 12
More intervention was not more control. The successful policy acted early, read the response, and mostly stopped. The failed blind policy kept applying the strongest available correction.
04 / A hidden state, not a crude shape change

The causal worlds diverged while coarse field displacement stayed almost identical.

Open and closed branches were both far from the untouched state. But their distance from untouched was nearly the same—the mean open-minus-closed field-distance contrast stayed near zero.

This is not simply “one controller moved more pixels.” The intervention changed what the state could still become under alternative histories.

Open minus closed field displacement

P4: -0.000   while causal separation was +0.037

P6: +0.013   while causal separation was +0.029

P8: +0.012   while causal separation was +0.027

P12: +0.000   while causal separation was +0.018

05 / What this means

Developmental commitment is now something we can control.

P1strong state-setting pulse
P2–4feedback mostly chooses restraint
Releaseall control stops
P6–8large causal memory remains
P12memory is smaller, still unmistakable

In plain words: early Flow Lenia development contains a controllable gate. A correctly timed pulse can place the same material into a state that remains more responsive to its later causal history. Repeated brute-force correction closes that gate instead.

06 / Breadth

This was not one lucky creature.

g010.049
g020.066
g030.075
g040.045
g050.100
g060.052
g070.036
g080.100
g090.095
g100.068
g110.039
g120.068
g130.082
g140.117
g150.122
g160.089
07 / The next direct swing

Use the open gate to rewrite what organism forms.

Visible transition timing did not yet separate cleanly: open minus closed advance +4.3 steps, interval [-8.8, 18.1]. That is the next frontier, not a reason to shrink the present finding.