The score came back.
The future had moved.
After eight free steps, adaptive feedback rebuilt a huge information split in every organism. But that split no longer narrowed the independent future repertoire. The developmental direction had separated from the instantaneous information gradient.
Drive. Release. Reacquire.
Track the moving direction
Four adaptive turns create the q48 split and narrow the microscopic future.
Release for eight steps
The local split disappears. At q57, all eight actions are tested again from the released state.
Reacquire the score direction
High chooses the new maximum; low chooses the new minimum. The q60 split is +1.607.
Challenge independently
Nine matched futures ask whether the reacquired state has fewer ways to continue.
The q48 split vanishes in one step.
This is not a slowly draining reservoir. Whole-minus-sum is a local directional quantity: the deliberately selected +1.393 split at q48 becomes −0.351 at q49, then wanders around zero.
The useful action rotates again.
The controller can find a new information direction at q57. It is genuinely state-dependent, and it recreates a clean q60 split.
Reacquiring the score does not reacquire the future.
At q60, maximizing the local information score no longer makes the high branch’s future narrower. Compared with the stale q45 action, it makes the high−low future contrast +0.008 wider, 95% CI [+0.001, +0.016].
The stale effect is not carried by one animal: every leave-one-ancestor mean stays negative. Six of eight organism groups point negative. This was a predeclared same-mass control, not a contrast invented after seeing the result.
The two directions align only at certain ages.
Local information direction and future narrowing align.
Tracking the rotating direction recovers almost the same contraction.
The score can still be controlled, but it no longer points toward fewer futures.
This is the new object: not a universal scalar called “commitment,” but a developmental window in which an information gradient aligns with the organism’s future-constraining direction.
Same released organism, three q57 choices.
Representative: harmonic-dancer-0397. High branch left, low branch right.
Before rescue.
New score extrema.
Old direction replayed.
Free continuation.
Find the alignment window.
At a sequence of developmental ages, independently re-estimate the local information gradient, perturb high and low, then measure the untouched future repertoire. The target is a temporal map of where information geometry actually constrains development—and where it becomes merely a controllable readout.
We have a candidate signal of developmental commitment: a phase where whole-over-parts information and loss of future possibilities become causally aligned before the organism’s visible transition.
Then we learn a state-dependent detector of readiness rather than a global clock—and can ask whether its peak predicts each organism’s own morphological turn.