The model’s kernel has three readouts — three heads sharing one recurrent core, each answering one question a growing tissue must answer at every moment.
The differentiation head — what. Cell identity is a softmax over super-enhancer attention: each candidate fate is a cluster of master-transcription-factor super-enhancers, and commitment is the entropy of that distribution collapsing. The head is gated by the PRC2–TERRA clock — a fate can only be chosen once Polycomb repression, withdrawn on the clock, falls below that fate’s threshold. Validated against the zebrafish atlas: the model reproduces the order in which cell fates appear (Spearman 0.95).
The division head — when and how much. Proliferation drives growth — and drives the clock itself, since each division shortens the telomere. Division is therefore the escapement of the whole system: the zebrafish embryo’s cell count follows an open-ended power law (N ≈ 132·t1.91, R² = 0.98 against measured counts), not a saturating curve, and every tick moves every cell’s clock forward.
The motility head — where. Cells move: convergent extension narrows and lengthens the body axis, collective migrations carry primordia to their addresses on the bioelectric frame, and the same head’s failure modes are the classic birth defects (a fold that does not close is a neural tube defect). Shape is not painted — it is the mechanical consequence of motility executed under the other two heads’ instructions.
All three heads are read from the same genome by the outer model and integrated by the same inner automaton — see the NCA+LGM model. Grown together from a single cell, they produce a recognizable vertebrate body.