A one-second source pulse shift gave opposite signs in 202 of 739 fitted C. elegans transfer models under a 10% kernel change bound

We tested whether one sign can describe a fitted calcium transfer model for two source inputs with the same height, duration and area. We used 739 selected wild-type C. elegans kernels from a pinned signal propagation atlas. Each had q below 0.05, at least three finite stored response values and a nonzero kernel. We shifted a one-second source calcium pulse by one second and read both model responses at two seconds. The signs differed in 268 kernels. In 202 kernels, both signs remained for every kernel change with total absolute area below 10% of the original absolute kernel area on zero to two seconds. We prove that this change bound is sharp for arbitrary integrable changes. Smooth pulses kept opposite signs in all 202 cases. The same area rule gave 3 cases with half-second pulses and 238 with two-second pulses. Of 198 directions with a direct link in at least one of four native anatomical maps, 43 had the main property. We keep all 302 canonical neuron slots for route checks. A signed-walk argument states a necessary condition at matched neural ports. An exact control also shows that one positive neural route and nonnegative cell filters can give a signed calcium transfer kernel after division by the source calcium filter. Thus these fitted signs do not identify inhibition or prove biological pulse sign reversals. The analysis is exploratory. The export lacks event-level kernel uncertainty and animal identities. High-precision integrals, separate numerical integration, source checks and a complete replay verify the calculations. The package preserves code, derived tables, source provenance, figures and repeat instructions.

connectome · signal propagation · calcium imaging · transfer kernel · pulse timing · response sign · source normalization · positive systems