Partial readouts gained hidden-input information without readout noise in two positive adult worm circuit models

We tested whether partial neural readouts must lose pulse information when full-state information decreases. In stationary linear Gaussian models without readout noise, information decreases for every input exactly when observed variables form a closed dynamical system. For positive couplings, a shortest path of d edges from a hidden input to the readout gives information onset proportional to time to power 2d. We tested binary and count weights, two gains and two name-based readouts in a common 157-cell core of two adult worm chemical maps. All hidden cells reached a readout in one or two edges. All 16 settings had rising hidden-input information, but mean grid peaks were only 0.065 to 0.360 percent of initial full-state information. Full-state upper bounds held. This applies established Gaussian and observability methods; it does not show new information creation. Two partial maps, positive coupling, fixed noise, equal leaks and no matched activity limit biological inference. Original code, a proof, aggregate results, checks and external retrieval steps support reproduction. External records are not bundled.

connectome · partial observation · Fisher information · hidden inputs · C. elegans · shortest paths · linear Gaussian model