Shared neuron terms gave no clear delay prediction gain in a fixed completion test of 1,529 mouse cortical connections

We tested whether delays on other connections of one neuron help predict a hidden connection delay. We read all 123,506 directed pair records in Allen Synaptic Physiology r2.1. The primary sample had 1,529 mouse cortical connections from 560 donor groups. Global fitting and model choice excluded each test donor. Other eligible delays in its recording served as calibration. A ridge reference used cell labels, distance, depth, and recording conditions. Local covariance terms represented the same experiment, source, and target. Adding source and target terms to the experiment rule reduced mean squared error from 0.272546 to 0.271057 ms squared. The decrease was 0.547%, with a conditional 95% donor resampling range of -0.574% to +1.556%. Planned checks in milliseconds also had ranges that included zero. A natural log delay check gave a 1.300% decrease, with a range of +0.012% to +2.495%. A separate 143-connection human sample gave no clear gain. We reconstructed all 2,665 finite source delays from accepted average fits. Direct masked solves checked every primary prediction. A full source repeat matched predictions, losses, and model choices. The package preserves new code, all results, figures, source provenance, a completion tool, a reader for the complete recording graphs, and repeat steps. Original source data remain external. These selected slice circuits are not complete anatomical connectomes. The results limit this additive completion model and do not establish independent delays or a causal source effect.

synaptic delay · cortical microcircuits · multipatch · connectome models · delay completion · mouse cortex · shared source · shared target