State split at q56
0.15695% bootstrap [0.148, 0.165] · 30/30 nonzeroThe result
The organism kept a memory of how it was driven.
Seven passages after the last pulse, alternating and blocked histories were already 0.156 Hellinger units apart. Later, under both free continuation and eight identical challenge seeds, that separation had grown to 0.341. Every ancestor showed the split.
Later future-cloud split
0.34195% bootstrap [0.311, 0.371]Amplification
2.25×95% bootstrap [2.08, 2.41]Cloud shift / cloud width
1.01×The displacement was about one entire repertoire widthWhat changed
Same counts. Same beginning. Same ending. Different syntax.
Each pair began from the exact same q32 organism. Within an ending direction, both branches received the same number of high and low score-seeking pulses and shared the same first and last direction. Only the order of the middle instructions changed.
Ending high
Three high-seeking and two low-seeking decisions.
AlternatingHLHLH
BlockedHHLLH
Ending low
Two high-seeking and three low-seeking decisions.
AlternatingLHLHL
BlockedLLHHL
Because this was closed-loop feedback, “high” and “low” selected actions from the organism’s current state. The physical seed sequence could therefore diverge. That is the point: we changed the temporal grammar of feedback, not merely replayed an open-loop seed list.
What kind of memory?
Location changed. Volume did not.
A causal repertoire has at least two distinct properties: how broad it is, and where it sits in form-space. Our frozen prediction expected alternating feedback to preserve more breadth. It did not. The larger discovery is that order translated the whole repertoire instead.
Breadth: nearly unchanged
→
Location: separated by one cloud-width
Alternating − blocked breadth
-0.004795% bootstrap [-0.0117, 0.0020] · 15 positive / 15 negativeDistance-matrix correlation
0.07795% bootstrap [0.024, 0.130] · causal geometry was largely rewrittenChallenge-rank correlation
0.06095% bootstrap [-0.040, 0.161] · weak action-response correspondenceSame strongest challenge
20%Only 12 of 60 ending-specific comparisonsOne representative ancestor
The split survives release.
These are the ending-high branches of harmonic-cell-0173 (source 12), selected as the median ancestor by immediate history separation—not as an extreme example. The left pair is shortly after the pulse train; the right pair is the later unchallenged continuation.
The new picture
The future has a position, not only a size.
Flow Lenia does not merely remember whether it was pushed “up” or “down.” It remembers the temporal arrangement of those pushes, and carries that history into the organization of its later possibilities.
What this adds
Our previous swing found that future-form breadth was already broad and fairly stable across age. This swing explains how commitment can still happen: the organism can move between different repertoires of similar size.
That is a cleaner model of developmental memory than a single scalar becoming larger or smaller. The causal landscape itself is being translated and weakly remapped.
Where this points
The next direct swings are obvious now.
- ReplayReplay the exact physical action sequences open-loop. If the split weakens, the memory belongs to closed-loop feedback; if it remains, the action history itself is sufficient.
- LoopDrive high → low → high and low → high → low through the same score range. A failure to retrace the same causal repertoire would be strict hysteresis.
- TranslateMeasure repertoire position continuously during development, not only repertoire width. We may find the state turning into a new basin before morphology visibly changes.
- BirthAsk whether a sudden repertoire translation precedes identity turnover, ecological succession, fission, or the appearance of an organism-like body.