Eight filter stages kept gain-limit error below 5% in delayed models of 91 Allen mouse E-I pairs

We tested replacement of fixed onset delays by first-order filter chains in negative feedback models of 91 selected Allen mouse visual-cortex excitatory-inhibitory pairs. Both measured voltage decay times stayed fixed. A one-stage chain with the same mean delay raised the critical gain by a median 26.40% and at most 65.01%. Four stages kept the error within 5% in 54 of 91 pairs. Eight stages did so in all 91 pairs, with a maximum error of 4.65%. The least stage count ranged from two to eight. Doubling delays removed the uniform eight-stage guarantee. Independent phase, cubic, selection, and complex-root checks passed. The delay-distribution effect is established theory; this work supplies a linked-data approximation benchmark. Voltage decay times are assumed model filters. No neuronal gains, brain stability, oscillations, behavior, or representation reliability were measured. Three complete human pairs do not support a transfer test. Code, extracted evidence, source hashes, sensitivity results, and offline reproduction are included.

Allen Institute · synaptic delay · feedback stability · filter chain · Erlang delay · model approximation