Reverse links gave no clear error decrease in a ridge test of response variability for 888 mouse cortical connections with test animal groups excluded

We tested whether a detected reverse connection helps predict variation among synaptic responses. We used Allen Synaptic Physiology r2.1 and 888 mouse cortical directions from 405 animal groups inferred from specimen names. A fixed ridge reference used cell labels, layers, distance, response size, noise and measurement counts. Adding reverse-link terms increased squared test error by 0.71%. Every test animal group was absent from transforms, penalty choice and fitting. Across 199 donor resamples with the full procedure refitted, the central 95% range for error decrease was -4.10% to +2.15%. New animal splits and added measurement checks also gave no clear gain. A separate exploratory test of 114 human directions from 36 donors gave no clear gain. The source has sampled physiological recording graphs, not a complete anatomical map. Only 888 of 1,175 otherwise eligible mouse directions had a finite outcome. Missing reverse connections, selected outcomes, uncertain donor identities and absent event fits limit inference. The result limits the tested prediction rule; it does not establish equal mechanisms or no biological role for reciprocity. Independent SQL and solver checks passed. A fresh run reproduced 115 numerical files exactly. The package preserves original code, plans, source versions and hashes, fitted models, test results, figures and repeat instructions. Original source measurements remain external under their terms.

synaptic variability · reciprocal connections · cortical circuits · structure and function · grouped prediction · Allen Synaptic Physiology · reproducibility