Tier 1
CPU/CI robustness, survival, human-utility and deterministic attack coverage where applicable.
notdataMEASURED CLAIMS ONLY — unexecuted compute remains NOT_TESTED
Research & Benchmarks
NOTDATA separates measured evidence from research intent. CPU-first Tier 1 can measure survival and human utility; Training Utility Gap and training-trace claims require valid model experiments.
Tier 2
Bring-your-own-compute paired model experiments with precommitted datasets, seeds, metrics and thresholds.
Tier 3
Funded or collaborative campaigns for broader model families, adaptive attacks and replication.