We were asking
Can adaptive feedback follow the rotating information landscape better than replaying the action that worked earlier?
Flow Lenia / control
The Controller Turns — and the Future Narrows Again
A state-aware controller re-tested eight actions at every pulse, repeatedly changed its choice, rebuilt the high-versus-low information split, and recovered later future contraction without selecting on that future measurement.
libraryadaptive closed-loop intervention
This experiment found
Yes; action rankings re-indexed almost completely, and adaptive selection restored both the local split and an independent contraction of microscopic futures.
That sent us to
Does the controlled state preserve that alignment after release, or must the controller keep following it?