Experimental study demonstrates metabolic gating and scarcity pricing in gradient-free learning cells, suggesting physiological bounds govern memory durability.
Companion to "Tree Memory: An Interpretable, Gradient-Free Learning Stack and Its Measured Laws of Memory, Value, Agency, and Governance" (2026). Learning machines are built from memory and plasticity, but not from metabolism. This paper constructs and measures the stress physiology of a previously certified, fully inspectable learning cell: hunger as a broadcast signal entering at a receptor port; metabolism as a per-operation energy meter whose refusal floor cannot be overridden; a neuromodulatory gate deciding which experiences are written (binary admission, graded depth); and their crosstalk — scarcity prices plasticity. The instrument is scaled to a two-cell population sharing one economy, measuring correlated forgetting, history-diversity rescue, an observer cell's knowledge of the colony's own gap, a first-draw privilege in shared scarcity, and its measured remedy, rationing. Across twenty campaign families conducted under a pre-registration discipline — predictions declared before data, enforced kill conditions, derived (never tuned) doses, byte-for-byte determinism across machines — the paper measures a body of laws: desperation vigilance; starvation that silences commentary before memory, with survival governed by the price ladder; a write gate that dials continuously with signal strength; a famine band whose threshold tracks the reserve with a measured midpoint; a gate that prices the present until a history channel makes duration — not depth — the chronic variable; scars that saturate, fade on two clocks, and remember how long the famine lasted; and fairness as the measured remedy for a commons. Several registered predictions failed and are reported as such, each with a located mechanism. Every component is standard-library only and deterministic; all campaign harnesses carry their registrations, gates, and kills in their headers; an append-only record documents every campaign, amendment, and verdict. Same seeds, same streams, same numbers — verified across independent machines. All code, harnesses, and the complete campaign record accompany the paper. The work is positioned as an instrument with a published physics, not a theory of mind: every biological analog is paired with its measurement, and every measurement with its boundary.
No takes yet. Share an insight, caveat, or question.
Darius Puzinas (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: