Code, protocols and data for: Null-model treatment of the sensory-motor boundary changes an evolutionary connectome comparison

Abstract

Code, frozen pre-registration protocols, analysis outputs, figures and manuscripts for the preprint Null-model treatment of the sensory-motor boundary changes an evolutionary connectome comparison. Agents whose brain is a compressed adult Drosophila FlyWire v783 connectome evolve in a 2D foraging world next to populations built on randomised wiring. Two standard randomisations move about a tenth of olfactory receptor output directly onto descending motor groups, against 0.012 % in the compressed connectome. Controls that preserve the sensory and motor boundary and randomise only the interior, replacing about nine in ten interior edges, give seed-mean differences inside a ±0.10 equivalence bound. A graded shortcut transplant raises fitness in step with its size, and against a sham that rewires about as much (101 swaps against 68 to 129) the full dose is ahead in 10 of 10 seeds, including every seed in which it rewired less. New in version 4.1. Corrections from a line-by-line audit of every number in the manuscript against its source; none changes a conclusion. The dose sizes and swap counts in section 3.5 were seed-0 values and are now ranges over seeds; the full dose and the sham rewire comparable rather than equal amounts, and the per-seed data show this strengthens the comparison. The interior overlap of the boundary-preserving nulls, a confidence-interval statement in section 3.1 and the Figure 9 caption are corrected. make_paper_numbers.py regenerates values that version 4.0 quoted without a saved source. The repository also carries the protocols of follow-up experiments not reported in the paper. Contents. flyconnectome-nulls-v4.1.tar.gz: engines, analysis and figure scripts, protocols, results, research log (snapshot of the public GitHub repository). preprint.pdf, supplementary.pdf, alife_short_version.pdf: manuscripts. Not included. Raw per-generation logs and stored genomes (about 40 GB); the analysis outputs computed from them are in results/. FlyWire v783 connectivity and annotations are public and are not redistributed. Drafts and results were reviewed by language models as AI-assisted review, not peer review.