Constrained Neuroevolution - code, full evolved-agent runs, and per-frame activations

Code and full evolutionary archive backing Khetan & Asopa, "Selection navigates a degenerate circuit space: behavioral individuation without structural differentiation in constrained neuroevolution." constrained-neuroevolution-code-v1.zip (~76 MB) — frozen snapshot of the project repository at the commit corresponding to this manuscript: simulation and evolution code, analysis scripts, the distilled best-evolved agents, the analysis intermediates, and the manuscript source. This zip alone reproduces every figure and statistic in the paper: unzip, pip install -r requirements.txt, and run rebuild_all.sh. The exact commit is recorded in ARCHIVE_COMMIT.txt inside the zip. The live repository is at https://github.com/pranetkhetan/constrained-neuroevolution. constrained-neuroevolution-data-v1.zip (~11 GB) — the full evolutionary archive: all 150 generations of 54 specialist runs (9 mice × 6 replicates) and 15 generalist runs, plus the full per-frame activation tensors for all 54 specialist agents. Not required to reproduce the published results; use it only to re-derive the analysis intermediates from scratch or to audit the evolutionary trajectories directly. Unzip and point NEUROEVO_DATA_DIR / NEUROEVO_ANALYSIS_DIR at its data/ and analysis/ subdirectories (see the repository README). Raw mouse behavioral data are not included: they are the tracked-trajectory files of Rosenberg et al. (2021), eLife, doi:10.7554/eLife.66175, which we do not redistribute. Code is released under the MIT License (see LICENSE in the code zip).