Local link scores captured 85 to 86 percent of the top removal effect in two worm noise models at small leak

We tested whether small-change link scores identify full removals that most increase stationary variance in two passive adult worm electrical models. At leak equal to 0.01 times mean weighted degree, the local top lists captured 85.08 and 86.46 percent of the summed effect in exact top lists. Rank correlations were 0.9509 and 0.9629. Yet individual variance increases exceeded local scores by factors up to 63.81 and 92.87. At leak ratio 1, capture rose above 96 percent and the largest correction fell below 2.23. Unit link weights gave 75.49 and 100 percent capture at small leak, despite large size errors. A standard rank-one inverse identity gave exact full-loss effects. All 3,312 cases passed full-spectrum checks, and 48 passed independent covariance solves. The contribution is a reproducible benchmark and a test of leak and outcome dependence, not a new matrix identity. Incomplete electrical maps, only two adults, and assumed passive parameters prevent claims about measured neural noise or reliable computation.

C. elegans · gap junction · passive noise · link removal · membrane leak · network coherence