Adaptive Flow Lenia control

The controller turns—and the future narrows again.

The useful action changes almost every time the organism changes. A controller that re-reads the state rebuilds the information split; blind replay cannot.

The downstream microscopic future contracts again—even though future repertoire was never part of the action-selection rule.

+1.393adaptive high−low WMS at q48 · all 30 positive
-0.008independent future-repertoire contraction
-1.443loss after eight unforced steps
01 · THE CONTROLLERS

One repeats. One listens.

Open-loop replay

Ask once, repeat forever.

1A62A63A64A6

The action selected at q33 is replayed at q37, q41, and q45 regardless of what the organism becomes.

VERSUS
Adaptive re-selection

Ask again at every turn.

1A62A23A34A2

At every pulse, test all eight actions from the current state and select the one that now points high or low.

For the representative organism, the high controller uses A6, A2, A3, A2; the low controller uses A1, A4, A5, A7. No global “high action” exists.

02 · THE FIELD ROTATES

The best action re-indexes almost completely.

93%high controllers change action at turn 2
97%low controllers change action at turn 3
3.5distinct actions in an average four-turn sequence

The rotating direction is not a metaphor. Between 83% and 97% of later decisions differ from the action that was optimal at the first pulse.

03 · CONTROL

Adaptive tracking rebuilds the split.

adaptive high−low
+1.393 [+1.035, +1.800]
replay high−low
+0.408 [-0.049, +0.901]
adaptive advantage
+0.985 [+0.506, +1.493]

At q48, every one of the 30 adaptive high branches is above its adaptive low sibling. The adaptive advantage over replay is +0.985, 95% CI [+0.506, +1.493].

04 · THE INDEPENDENT CONSEQUENCE

The future narrows again.

Original one-push window · q36
-0.008

High-commitment siblings had fewer microscopic futures.

Adaptive four-turn control · q48
-0.008

The contraction returns at nearly the same cohort magnitude.

adaptive high−low repertoire
-0.008 [-0.017, +0.001]
replay high−low repertoire
-0.001 [-0.009, +0.007]
adaptive − replay
-0.007 [-0.017, +0.004]

The high adaptive state averages 0.333 pairwise Hellinger across its nine challenge futures; the low state averages 0.340. Seven of eight organism groups point negative or exactly zero, and every leave-one-ancestor mean remains negative.

The 95% interval just touches zero, and the size of an organism’s WMS gap does not predict its individual repertoire effect (Spearman +0.074). The recurrent result is at the controlled cohort level, not a simple per-organism dose law.

05 · VISIBLE STATES

Same ancestor, different control histories.

Representative: serene-form-0965. High on the left, low on the right.

adaptive high q48adaptive low q48
Adaptive · q48
Four state-dependent turns.
replay high q48replay low q48
Replay · q48
The first action repeated four times.
adaptive high q56adaptive low q56
Adaptive released · q56
Eight unforced steps later.
06 · MEMORY

The state can be held. It does not yet hold itself.

q48 · active control
+1.393high−low WMS
→ 8 free steps →
q56 · released
-0.050high−low WMS

The split falls by -1.443, with 29 of 30 organisms losing separation. The field-repertoire difference remains directionally negative (-0.005) but is not cleanly resolved.

NEXT DIRECT SWING

Map the maintenance law.

Pulse frequency

Re-select every 2, 4, or 8 steps. Find the minimum feedback rate that holds the split, and whether the required rate changes near morphological transition.

Hold → release → rescue

Drive the state high, release it until the split decays, then resume adaptive control. Test hysteresis, recovery speed, and whether a formerly driven organism becomes easier to steer.

That turns our new qualitative discovery into a dynamical law: the bandwidth, memory, and reversibility of the information-geometric control field.