Code, data and figure sources for "Matter that learns what matters"

Code, data and figure sources for the article “Matter that learns what matters” (N. G. Berloff, DAMTP, University of Cambridge) and its Supplementary Information. The deposit contains the simulation code and its separately written Python implementation, the scorers, the run seeds, configurations and source pins, per-cycle histories and scored criterion labels for every network study in the paper, the design records fixed before the runs, the analytical check scripts, the retrospective reduced stochastic model, and the scripts and data for all article and Supplementary figures, with one script that regenerates them. Within-cycle observer records are not included; they are regenerated from the code and seeds. Each top-level folder is a separate archive; README.md maps each section and figure of the paper to its location, and SHA256SUMS.txt lists checksums of all files. Requirements: Python 3 with numpy, scipy, matplotlib, pandas and numba; a C compiler for the simulation code. About 1.8 GB.