Input-Linked Formation and Selective Removal in an Online Cognitive Runtime: A Traceable Attractor-Population Lifecycle — Data and Code

Anonymous data and code archive for the manuscript "Input-Linked Formation and Selective Removal in an Online Cognitive Runtime: A Traceable Attractor-Population Lifecycle" under double-anonymous peer review at Neural Computing and Applications. The archive contains the cognitive runtime source (24 Python modules under runtime/), the manuscript-facing reproducibility package of 978 per-run JSON summaries together with frozen prompt pools, embedding-space geometry, and the factorial-intervention aggregates (raw_data/), the original per-seed runtime state for the six paper-5D experiments covering formation, pruning, the decay-by-ecology factorial intervention, second-family replication, and the 16-family boundary map (experiments/), and the five matplotlib generator scripts that produce Figures 1 to 5 of the manuscript (figures/). A top-level README.md describes the directory layout, and DATA_INVENTORY.md provides a claim-to-file mapping for every manuscript section. All formal experiments use a fixed sequence of 30 prompt-order seeds (101, 202, 303, ..., 3030), and every numerical claim in the manuscript is recomputable from the corresponding files documented in DATA_INVENTORY.md. The archive has been scrubbed of identifying absolute paths and signatures for double-anonymous review; author and affiliation metadata will be added on acceptance.