Before the Replicator / Direct causal swing 01
Future-blind Flow Lenia intervention

The future remembers the first push.

A tiny early perturbation chosen by information geometry changed how strongly a later intervention survived being followed by another intervention.

Φ-forward increased later action memory by 0.00392. The 90% family-bootstrap interval was +0.00044 to +0.00781. That is a 9.5% increase over the Φ-backward mean.
Same early organism · same age
Small local swap selected to move the state Φ-forward
↔
Matched swap selected to move the state Φ-backward
Same future randomness
same later action A
same later action B
Φ-forwardA remains visibleafter B arrives
Φ-backwardA fades moreafter B arrives
2,432future trajectories
19organic families
152complete 4×4 grammars
29,184decoded future fields
01 / The question

Can invisible organization steer development?

We stopped asking whether an information score merely rises before something interesting. We pushed the same early organism in opposite information-defined directions, then asked whether the meaning of later actions changed.

1

Choose without seeing the future

Eight matched local swaps were scored from a short prefix. The Φ-forward and Φ-backward swaps were selected before any later action sequence was generated.

2

Give every state the same future

Every treatment received the same random future and every ordered pair of four founder actions: 4 first actions × 4 second actions.

3

Ask what survived

Holding the second action fixed, we measured how different the later fields remained when only the first action changed. That is causal memory: how much history survives being overwritten.

02 / The finding

The early push changed overwrite resistance.

Φ-forward did not simply make the creature “more complex.” It selectively increased the field-level trace left by the first later action after the second action arrived.

+9.5%

more first-action memory than Φ-backward

The forward mean was 0.04523; the backward mean was 0.04131. Across the 38 family-schedule states, 21 favored forward, 13 favored backward, and the four zero-leverage states were exactly tied.

The same later action B overwrote less of action A when the organism had first been nudged in the Φ-forward direction.
The separation is not an immediate artifact. It accumulates through development: near zero at passage 1, +0.00347 by passage 6, and +0.00500 by passage 12.
03 / The shape of the effect

Memory moved. The whole grammar did not.

This was not a generic increase in every measure. The information-directed perturbation found a specific axis of causal plasticity: whether developmental history remains readable after subsequent input.

Φ-forward minus Φ-backward. Only first-action field memory separated cleanly in the frozen panel. Intervals are family-bootstrap 90% intervals.
First-action field memoryDoes A remain visible after B?
+0.00392
[+0.00044, +0.00781]
Whole causal-grammar rewriteDoes B mean different things after different A?
+0.00089
[−0.00071, +0.00268]
Second-action breadthHow far apart are the four B outcomes?
+0.00081
[−0.00112, +0.00270]
Composition-grammar rewriteDid founder shares acquire different action geometry?
+0.00001
[−0.00157, +0.00167]

The forward–backward memory contrast was precomputed in the future-blind analysis. Forward versus untouched was +0.00195 and forward versus the blind swap was +0.00365, but those separate post-opening intervals crossed zero. The clean statement is directional: opposite information-selected perturbations changed later field memory.

04 / Across organisms

Real, heterogeneous, and not a single-family trick.

The average effect is not universal. Some organisms reverse it. That heterogeneity is useful: it gives us a map of which developmental states can convert an information push into memory.

Each point is one family, averaged over its two fresh schedules. Ten family means favored Φ-forward and nine favored Φ-backward; the positive average comes from larger forward effects in several families, not from unanimous sign agreement.
Built-in causal sanity check0.000000maximum difference when forward and backward selected the same swap
05 / Why this is causal

When the push collapsed, the future collapsed too.

Four family-schedule states had zero information leverage: the forward and backward selectors chose the exact same local swap. Across every later field metric and every passage, those paired trajectories were exactly identical.

That is unusually clean. It shows that treatment labels, output paths, and analysis bookkeeping did not manufacture the effect. A difference appeared only when the early physical state was actually pushed in a different direction.

06 / What opened

Information geometry may tune causal memory.

The early state did not choose the organism’s final form outright. It changed how much later history the developing system retained. That is a direct bridge between information geometry and developmental commitment.