The Fly Brain Against the Cortical Column: A Second-Substrate Audit of the Computational-Primes Table for Neural Tissue

A compiler mapping operations onto tissue needs a cost model: per operation, the condition under which tissue runs it. A companion paper (10.5281/zenodo.23218412) built it from mammalian cortex for twelve computational primes (elementary operations such as summation) and four residue capabilities (such as addressing, which a finite circuit lacks). We hold it against a second substrate, the adult Drosophila. On two connectomes and published physiology, 9 of 32 scored comparisons hold, 4 are sharper, 1 falls, 6 are not decisive and 12 not applicable. The contrast law is not decisive: its pre-registered measure, which includes an assumed frozen weight scatter, gives x*_70 = 3.38 at the ideal reader in a 1 s window, above the fall bound 2, but 1.37 in a 100 ms window, which is not a fall. One row falls, in form: fly synapse counts follow a power law (exponent −1.67), not the cortical log-normal; whether one connection is written less precisely than in cortex is not decisive, because its left–right count scatter lies inside the cortical within-connection band. Addressing is not applicable, and the fly adds a different mechanism: it addresses through wiring (15 compartments, each with a write port), whereas the cortical row counts items held electrically; the fly has no number for that bound, so it neither confirms nor contradicts that the bound is mammal-specific. Rows resting on a governing equation or structural statement transfer, e.g. the membrane read-out and the absence of a detectable positive correlation between same-type axons on one dendrite. The fly adds blueprints: a dopaminergic teacher as write port, one reader per compartment, a subtractive adaptive threshold in the model, and a heading ring with exactly one wiring loop in the single-hemisphere connectome. In no tested case does its wiring alone fix a time constant: seconds-long persistence is measured, and a linear ring model holds 30 s only if one inhibitory weight is tuned within 1%.