Nonnegative cell delays fail in seven forest groups in selected Allen mouse slices

We tested whether stored synaptic delays equal a nonnegative source-cell term plus a nonnegative target-cell term in Allen cortical slice data. Seven selected mouse groups with no graph cycles fail an exact fit, although unrestricted cell terms fit them exactly. Six remain after a 0.2 to 5 ms delay filter. One remains with strict cell labels and with both filters. The largest required uniform correction in the primary sample is 0.609 ms, but the interior filter removes that case. The strict-label forest needs 0.225 ms. No added human group fails. Linear programming and independent graph certificates agree; 1,022 small-graph comparisons pass. Positive cell terms are an extra assumption, and the latency reference does not ensure that assumption. Missing delays, uncertain fit error, selected slices and sparse human graphs prevent a biological or cross-species claim. Original code, aggregate results, a figure and external-data regeneration steps support reproduction. External records are not bundled.

Allen Institute · synaptic delay · nonnegative cell terms · sparse circuits · structure and function · bounded error