Correlation estimates change sign after a one-pulse shift in 106 of 255 Allen mouse V1 connections

We tested how pulse position affects correlation estimates in selected Allen mouse V1 slice connections. Trial fits reproduced 267 of 268 finite stored first-pair correlations within 1e-10. One inspected public source revision instead reproduced the later pair in all 267 cases with sufficient fits. The exact release build is unknown. On identical three-pulse trials from 255 connections and 82 inferred mice, the first and later pair estimates had opposite signs in 106 connections. The donor-weighted mean difference was -0.04337, with a nominal 95 percent interval of -0.07913 to -0.00637. The excitatory subgroup interval included zero. Most connections had about 25 trials. Only two opposite-sign pairs had both nominal Fisher intervals exclude zero. Thus, sign changes in sample estimates are not secure biological classifications. The result supports a trial-based pulse-position check before summary comparisons. It does not identify a defect in the stored correlation field, a release mechanism, or a rule across regions and species. We provide code, sufficient statistics, source hashes, replay checks, and retrieval instructions.

Allen Institute · synaptic response correlation · pulse position · reproducibility · circuit reliability · source replay