Twofold variance bounds fail a common-filter test in 12 of 22 screened Allen mouse input groups

We tested a necessary condition for one common current-to-voltage filter in Allen cortical slice circuits. A normalized nonnegative filter must add the same nonnegative temporal variance to every input of one target and sign. We derived an exact least multiplicative variance tolerance and applied it to resting response fits. In mouse V1 coarse matrix, 47 of 62 selected groups fail when each current and voltage variance may change by a factor of two. Removal of fits near expected parameter bounds leaves 12 failures among 22 groups with 47 inputs. Human coarse matrix leaves 6 failures among 13 screened groups. Other mouse series show lower failure fractions after screening. These are deterministic tests of fitted response components, not confidence statements or a rejection of passive filtering in real neurons. Tail extrapolation, clamp-state differences, fit quality, and selected slices limit inference. New code, proof, aggregate results, source hashes, and external regeneration instructions support reproduction.

Allen Institute · synaptic filtering · convergent inputs · waveform variance · model audit