The hidden state changed the direction of development.
At step 32 the open state was actually smaller. Around step 80 the trajectories crossed. By step 256, under the same actions, the open body was larger, longer, and more skeletal.
Open minus closed radius, same actions
The mean crosses zero near step 80 and becomes clearly positive from step 112 onward.
What was held constant
For each of eight morphology directions, we chose an action pair from the complete 64-action map. Then we replayed that exact pair into both hidden states and watched 224 unseen steps.
The action pair is identical. The only difference is whether the early developmental gate had been held open or driven closed.
At step 256 the raw radius difference is +0.756 [+0.422, +1.094] pixels.
The state turns a pulse into a body plan.
These are the same two action seeds in the same organism. Source 50 was selected mechanically as the median positive same-action shift for the expansive–skeletal–elongated direction.
Step 32
Step 96
Step 256
By step 256, the open state had a distinct morphology.
Larger. Less box-filling. More stretched along a dominant axis. The change is not a loose impression from a few frames; all three measurements use the same action-controlled paired panel.
The gate did not add generic randomness. It favored a kind of body.
Eight requested body directions
Each bar asks where the same action ended up at step 256. Positive means the open state moved farther toward that requested direction; negative means the closed state did.
The strongest clean direction was expansive, skeletal, elongated: +3.855 [+0.169, +8.111] early-map standard deviations.
That is a developmental polarity, not “more of everything.” Open trajectories increasingly expanded into long, sparse bodies; closed trajectories leaned toward compact, solid, round forms.
The action grammar faded. The developmental mode remained.
Advantage of matching action to hidden state
At step 32, state-matched action selection had a 1.13-SD advantage. It was still visible at steps 48 and 64, then largely washed out.
What this means
Early on, the same pulse has a different local meaning depending on the hidden state. Later, the details of which pulse was chosen matter less—but the state has already committed the system to a different developmental trajectory.
The action starts the trajectory. The state decides what kind of body the trajectory becomes.
We can now connect information geometry to a visible body without hand-waving.
This is the bridge we were after: a hidden, manipulable informational configuration precedes form and causally biases what sort of organism develops under the same later interaction.
Now hit the turning point directly.
Reopen or close the gate around steps 64–96 and see whether the body trajectory turns with it, overshoots, or shows hysteresis.
Ask feedback to hold a requested body direction after the early action grammar fades. Can control overcome the basin, or does the basin win?
Use the emerging expansion window to search for daughter bodies that separate, persist, and continue their own dynamics after release.