Gate-conditioned formation · 256-step continuation

The same later actions left two different forms.

Flow Lenia is a continuous field whose bright mass gathers into moving, deformable forms; an earlier feedback experiment produced two versions of each organism, an open branch chosen to preserve separation between alternative causal histories and a closed branch chosen to reduce it, and when we applied the same pair of later actions to both branches, their shapes remained far apart 256 steps later, although the particular action pair chosen to maximize the early split was no better at maintaining it by the end than the within-organism control.

What we wanted to knowWould the two gate-conditioned futures remain visibly different as the organisms continued to form?

Grayscale Flow Lenia field for the open branch of organism s63 at step 32, showing a compact diagonally striped form against black — The montage follows serene-spiral-0219 (s63, family g16), the same trace used in the sealed v1 report; the curves and numerical estimates below include all 39 organisms.

Grayscale Flow Lenia field for the closed branch of organism s63 at step 32, showing a compact diagonally striped form against black — The montage follows serene-spiral-0219 (s63, family g16), the same trace used in the sealed v1 report; the curves and numerical estimates below include all 39 organisms.

What remained

Under matched later actions, the open and closed fields ended about 0.45 Hellinger units apart.

What faded

The max-selected pair’s extra separation fell from +0.040 at step 32 to +0.002 at step 256.

Cohort39

organisms from 16 source families

Continuation312

matched trajectories

Readout79,872

simulator frames

01 · Freezing the continuation

At step 32 we kept the widest and narrowest open–closed splits, then watched another 224 steps without choosing again.

For every organism, each ordered action pair was applied once to its open branch and once to its closed counterpart, keeping the later causes matched. We retained the pair with the largest open–closed Hellinger distance at step 32, together with the pair that produced the smallest distance as a within-organism control; both choices and their action schedules were fixed before steps 33 through 256 were run, so the later morphology could not influence what had been selected.

01The earlier controllerFeedback prepared open and closed gate states, alongside untouched and blind-full comparisons.

02Matched replayEach action pair was replayed on both gate states, keeping the later causes the same.

03The step-32 choiceOne pair gave each organism its widest split, while another gave it the narrowest.

04The unseen continuationOnly after both pairs were fixed did steps 33 through 256 run.

Why keep two pairs?

The wider pair showed what happened to an intentionally selected early split, while the narrower pair showed whether an ordinary action pair would also separate after passing through different gate states; because the wider pair was chosen using step 32 itself, its extra separation would have to remain larger later, rather than being inferred merely because its two branches were different.

02 · Following both curves

By step 256, both action pairs left their open and closed fields about 0.45 apart.

Open–closed distance under both pairs

A Hellinger distance of 0 means that the normalized fields are identical; larger values mean less overlap.

widest pair at step 32narrowest-pair control

0.448The pair chosen for the widest early split ended here, with bootstrap interval [0.423, 0.472], while the narrowest-pair control ended almost alongside it at 0.446.

The extra separation selected at step 32 dwindled

This curve subtracts the narrowest-pair control from the widest-pair open–closed distance.

widest minus narrowest pairno extra separation

+0.002By step 256 the difference had reached this value, interval [-0.013, +0.017], while the 16 source-family means divided into 9 positive and 7 negative directions.

Read together

The different gate histories continued to produce visibly different mature fields under matched later actions, whereas the action pair selected to enlarge the early split no longer supplied extra leverage; what lasted was the divergence between branches, not control of a particular mature target.

03 · Locating the visible difference

Open fields finished broader and less rectangular, while their component counts changed more often.

Across the final 32 frames, the open branch had a larger radius and lower rectangularity than the closed branch, and over the full trajectory its component count turned over more frequently. Because these readouts describe several related aspects of the same 39 organisms, without a multiplicity-adjusted confirmatory family, they locate where the visible difference appeared rather than establish a second primary result.

+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 · Reading the continuation

The history remained in the form after the chosen action stopped standing out.

By step 256, the open and closed fields were still about 0.45 Hellinger units apart under either action pair, but it no longer mattered whether we had chosen the pair with the largest or the smallest split at step 32. In that precise sense, the experiment found memory of the earlier gate rather than durable control of a particular morphological target: developmental history changed the field in which later actions operated, although we still cannot name a mature form in advance and reliably steer an organism toward it.

Nor did the older rectangularity-based event detector locate the difference in a single transition, since the open-minus-closed timing shift was +0.8 steps with interval [-13.1, +14.9]; in this continuation the history appeared across continuous geometry and topology, not as a cleanly earlier or later event.