The universality of push-pull motifs in agents

Regulation is forced to use two opposing flows — not one — whenever disturbances push both ways and each available flow can only act in one direction.

The core insight is almost embarrassingly simple once stated clearly. If the environment can shove a variable up or down, and every physical mechanism you have available only shoves in one direction, then you mathematically cannot hold that variable in a range with a single mechanism. You need at least one that pushes up and one that pulls down. This is not a design choice or an evolved preference — it is a theorem. The paper proves it (Theorem 1), proves the converse that one of each suffices (Theorem 2), and establishes that two is the minimum (Corollary 1). The proof is remarkably lean: it only needs to evaluate the system at the two boundary edges, not anywhere in between. No fancy dynamics required.

The "restoring force" objection — can't a single spring-like term $-\kappa(x - x^*)$ do the job? — gets a careful two-part answer. Mathematically, yes, that term is a two-signed flow in disguise: it pushes up below the set-point and pulls down above it. So it already contains both arms. But whether those two arms must be separately driven by distinct physical processes is not a theorem — it is a thermodynamic thesis. The paper argues (but does not prove) that in most real systems, elementary flows are powered by irreversible free-energy gradients that run one way: a kinase adds phosphate but cannot remove it, a heater warms but cannot cool. This one-directionality is the physical signature of dissipation, and it is what makes the theorem bite in practice. The reversible extreme — a hydrogen atom, a diamond lattice — genuinely has a single conservative potential doing both jobs, and the paper acknowledges this cleanly rather than papering over it.

The information-theoretic layer connects this to BCOM's Kolmogorov Theory framework. A regulator that cancels the predictable part of a structured disturbance opens a compression gap exactly equal to that disturbance's description length (Lemma 1). Among all regulators that achieve this gap, the universal prior — Occam's razor made precise — selects the one with the best world-model per bit of regulator complexity (Proposition 1). The push-pull pair is the minimal architecture that can cancel a two-sided disturbance, so it is the minimal entry in that preferred set. The paper is careful here: this is a selection statement, not a certificate that any given regulator contains a rich world-model. A clamp or attenuator can also open a gap cheaply, and the assumptions (A1–A5) that make the gap equal the description length explicitly exclude those routes.

What the paper ultimately proposes is a new ontological primitive for the KT framework: the push-pull motif as the minimal realization of a Good Algorithmic Regulator. It is the smallest thing that can hold a variable in a band against a two-sided world, it is the same architecture whether it is stabilizing (homeostasis) or oscillating (rhythm generation — same flows, different timing), and it sits at the root of what the framework means by persistence and agency. The cross-disciplinary survey spanning thermostats, kinase-phosphatase pairs, excitation-inhibition circuits, and predator-prey dynamics is illustrative rather than probative, and the paper says so explicitly — flagging economics and politics as loose analogies where the controlled variable is often ill-defined. The neural case study is the tightest: the paper cites companion work showing that essentially all neural mass models are the same push-pull loop with progressively more biological detail added, which is the motif's cross-substrate invariance demonstrated rigorously within one field.