A ten-mode wiring advantage for velocity decoding did not repeat in a second C. elegans sample

We tested whether ten smooth wiring modes retain more movement information than ten modes of a local distance graph. We used identified head-neuron activity from 40 freely moving C. elegans and repeated the test in eight animals from a second study. We used five separate time blocks, a 30-second training buffer, and nested ridge tuning. The mean wiring-minus-distance skill difference for velocity was 0.069 in the first sample (97.5% animal bootstrap interval 0.011 to 0.120). It was -0.022 in the second sample (-0.063 to 0.012). The head-bending differences were uncertain after adjustment for two outcomes. Wiring did not clearly beat ten-dimensional random projections for velocity in either sample. Training-only principal components gave higher head-bending skill in both samples. All ten-mode primary graph spaces had positive eigenvalue gaps. A five-mode check was withdrawn because the distance space was non-unique in 26 animals. The result does not support a repeated decoding advantage for these smooth wiring modes. The samples differ in recording length and experimental context, and come from one laboratory. The wiring and atlas positions come from other animals. Code, compact inputs, held-out predictions, source hashes, graph audits, and static figures support reproduction.

C. elegans · connectome · neural activity · behavior decoding · graph eigenmodes · brain geometry · blocked validation · replication · null result