Shared sources raised count-loss variance by at least 25 percent in 17,702 selected FlyWire targets under equal failure rates

We tested the cost of treating source neurons as separate failure units in each neuropil. We used proofread FlyWire v783 contact counts from one female fly. The primary sample had 61,223 intrinsic target neurons with at least 100 incoming contacts and at least two regions each carrying at least 10 percent. We compared whole-source failure with independent source-region failure. Each single-region loss distribution stayed the same. At a common source failure probability, the median total count-loss variance ratio was 1.0615. In 17,702 targets (28.91 percent), the ratio was at least 1.25. With arbitrary source probabilities from 0.01 to 0.10, the exact minimum ratio was at least 1.25 in 5,429 targets. Contact thresholds and region selection were tested. Independent formulas and simulations checked the calculations. This is a model audit of direct unsigned contact loss. Equal contact effects, independent source failures, incomplete contacts, atlas regions, and one brain limit its biological meaning. No activity or lesion data were tested. Code, sufficient statistics, bounds, source hashes and a figure support reproduction.

FlyWire · source identity · input loss · neuropils · variance · circuit reliability · connectomics