Keeping PD and LP identities reduced Brier forecast error by 1.1% in fixed logistic tests of 409 crab preparations

We tested whether known cell identity helps forecast loss of a regular rhythm in the crab pyloric circuit. PD belongs to the pacemaker group. LP is a follower with inhibitory feedback. We used 38,407 current regular bins from 409 preparations in Zenodo version 1. Each 20-second bin predicted two nonregular bins after a 20-second gap. Five outer test folds excluded complete preparations. Inner folds selected the model penalty. The planned comparison kept each metric pair unordered. Adding five named LP-minus-PD contrasts reduced equal-preparation Brier error from 0.062597 to 0.061934, a 1.06% decrease. Its conditional 95% preparation bootstrap interval was 0.29% to 1.90%. A second animal split gave a 0.97% decrease. An added control kept complete unnamed cell profiles with the same number of activity terms. Names reduced error by 1.09% against this control, with interval 0.17% to 2.13%. The gain did not apply equally to all preparations. A stricter three-bin outcome had an interval that included no gain. These intervals hold fitted predictions fixed. The original state map used the full source data. Thus, source label construction is not independent of the test animals. The result adds a reproducible forecast benchmark linked to known circuit roles. It does not identify a synaptic cause or validate forecasts in a new study. The package preserves compact data, source checks, code, all predictions, figures and repeat steps.

pyloric circuit · structure and function · neuron identity · circuit stability · forecasting · Cancer borealis · pacemaker · reproducible data