Raw training artifacts for "Benign Exclusion and Consensus-Test Limits in Heterogeneous Federated Learning"

Raw archive behind the manuscript's tables and checkpoint analyses: per-run records of the canonical campaign, the client-update matrices saved at rounds 0, 9 and 29, and the frozen source snapshots that produced them. Code, summary tables and the scripts that regenerate them are in the project repository, https://github.com/Bygokcen/fed-mdbscan-g; its archive/README.md maps each analysis script to the bundles it reads. 01_canonical_campaign_audit-v2.tar.gz: per-run records of every run group of the canonical campaign, the five failure records, recovery attempts, reports, frozen source and pip freeze. 02_five_step_checkpoints.tar.gz: five-step gate-replay checkpoints, 81 update matrices. 03_canonical_reexecution_checkpoints.tar.gz: re-executed canonical runs and control runs, 90 update matrices. 04_mechanism_and_development_studies.tar.gz: cutoff development, instrumentation, batch control, cluster replay (3 update matrices), forward round, step control and unclustered-policy studies. SHA256SUMS, raw_archive.sha256: SHA-256 of the bundles and of all 5,028 files inside them. Each update matrix is a NumPy array of shape (90, 159010), float32. See README.md for extraction and verification.