Reciprocal wiring gave no clear added late-train prediction value after early response control in two Allen slice samples

We tested whether reciprocal wiring improves late-train response predictions after early response measurements are known. Allen Synaptic Physiology r2.1 supplied 185 mouse visual-cortex and 79 human temporal-cortex excitatory connections from 79 and 29 inferred donors. Models held whole donors out of training. At ridge penalty 10, adding reciprocity increased donor mean squared error by 1.23% in mouse and 1.43% in human. Reference-minus-comparison gains were -0.000387 and -0.000227, with conditional 97.5% donor bootstrap intervals of [-0.000988, 0.000119] and [-0.000966, 0.000438]. No tested sensitivity variant gave a clear positive gain. Early response measurements gave large gains, but shared trials, baselines, and normalization limit their biological meaning. This test does not establish equal dynamics or a causal effect. Selected slice graphs, inferred donor labels, measurement error, and conditional uncertainty limit transfer. Code, source hashes, derived evidence, figures, and reproduction checks are supplied.

Allen Institute · reciprocal connectivity · short-term plasticity · donor validation · prediction · null result