Power-two rise filters lower gain limits in models of 91 selected Allen mouse E-I pairs

We tested how response rise shape changes a negative feedback model based on linked Allen Synaptic Physiology r2.1 records. Power-two rise kernels have an exact three-pole transform per edge. In 91 complete reciprocal excitatory-inhibitory pairs from mouse visual cortex, the full-to-decay-only critical gain ratio had median 0.459 and range 0.127 to 0.703. All 91 gain and crossing-frequency ratios were below one. An equal-mouse geometric mean gain ratio was 0.447, with a descriptive 95% donor bootstrap interval of 0.420 to 0.472. An interior-decay check retained 64 pairs and the same direction. Replacing the four added poles with their mean transit delay gave a median gain ratio of 0.962 relative to the rise model, but a worst ratio of 0.547. Three complete human pairs do not support a transfer test. Independent selection, integration and characteristic-root checks passed. Power two, unit-DC normalization, isolated linear loops, stored fit error and selected slice records limit inference. No neuronal gain, living-brain stability or representation outcome was measured. The package supplies code, derived evidence, source hashes and reproduction instructions.

synaptic physiology · feedback stability · response rise time · Allen Institute · circuit models