A causal state becomes visible

The gate is still visible 256 steps later.

Flow Lenia is a continuous field in which bright mass gathers into moving, deformable patterns. We placed each pattern into an open or closed causal state, gave the two branches matched later actions, and followed every frame: their mature forms remained far apart, although the action chosen to maximize that split lost its special advantage.

The frozen questionDo gate-conditioned differences in the future map grow into visible organismal transitions?

Frozen simulator frame for the open branch of organism s63 at step 32 — One frozen illustrative trace: serene-spiral-0219 (s63, family g16). These raw simulator frames were already used in the sealed v1 report; they are not a hand-picked substitute for the cohort result.

Frozen simulator frame for the closed branch of organism s63 at step 32 — One frozen illustrative trace: serene-spiral-0219 (s63, family g16). These raw simulator frames were already used in the sealed v1 report; they are not a hand-picked substitute for the cohort result.

Supported

Open and closed branches remained morphologically separated through the held-out future.

Not supported

The step-32 action target did not retain an extra advantage at step 256.

Cohort39

organisms, 16 source families

Surface312

trajectories

Readout79,872

source frames

01 · What was fixed before the future

The experiment separated a branch-memory question from a target-control question.

At step 32, after seeing only the first short future, we recorded the action pair that made each organism’s open and closed counterparts most different, along with the pair that made them least different. We then extended both pairs to step 256. That design can tell us whether the gate-conditioned branches stay apart, and whether the particular target chosen at step 32 continues to outperform its within-organism control.

01Set the gateEach source organism entered open, closed, untouched, and blind-full branches.

02Apply matched actionsThe open and closed counterpart received the same ordered pair of later actions.

03Select at step 32Keep the largest and smallest open–closed Hellinger split per organism.

04Open the holdoutSteps 64, 128, and 256 were unseen when those pairs were selected.

Interpretation boundary

The maximum-divergence pair is hypothesis-generating because it was chosen on the observed step-32 field. The minimum-divergence pair is the predeclared within-organism action control. Later persistence must therefore be read in two layers: branch divergence itself, and the max-minus-min selection interaction.

02 · What lasted

The branches stayed far apart; the selected target did not stay special.

Branch divergence persists

Hellinger distance: 0 means identical normalized fields; larger values mean less overlap.

max-selected pairmin-selected control

0.448mean open–closed distance at step 256 for the max-selected pair, bootstrap interval [0.423, 0.472]. The control pair was nearly the same: 0.446.

The early target advantage fades

Max-selected open–closed distance minus the min-selected control.

selection interactionzero

+0.002at step 256, interval [-0.013, +0.017]. Family means split 9 positive to 7 negative.

What the run shows

A gate-conditioned causal history leaves open and closed branches with visibly different mature fields under matched later actions.

What it does not show

It does not show that the action pair selected to maximize an early split keeps steering the mature form toward a persistent target.

03 · What became visible

The mature difference appears in scale, compactness, and topological turnover.

Across the final 32 frames, the open branch was broader and less rectangular than the closed branch, while its component count changed more often over the full trajectory. These were planned morphology readouts, but the frozen analysis marks them as descriptive; the intervals below should not be mistaken for a multiplicity-adjusted confirmatory family.

+0.747terminal radiusopen minus closed
bootstrap interval [+0.429, +1.108]

-0.028terminal rectangularityopen minus closed
bootstrap interval [-0.054, -0.003]

+11.7component-count turnoversopen minus closed
bootstrap interval [+2.9, +21.3]

04 · The useful conclusion

Commitment here is branch memory, not demonstrated target control.

The result is interesting precisely because the two statements come apart. An early gate can leave the same underlying Flow Lenia material in branches whose later shapes remain strongly distinguishable, even when the action pair selected to maximize their early separation no longer beats an ordinary within-organism control. The gate appears to alter the developmental context in which later actions unfold; this experiment does not yet show that we can name a mature form in advance and steer toward it.

The frozen visible-event detector also remained unresolved: open minus closed formation advance was +0.8 steps, interval [-13.1, +14.9]. The visible evidence lies in continuous geometry and topology, not in a cleanly shifted transition time.