A dependency-free, model-neutral evaluator for wrong-subject attribution, information-cutoff violations, duplicate-evidence weighting, stale state, unresolved-state loss, false completion, improper handoff, missing provenance, and loss of human authority. The repository includes synthetic cases, deterministic reports, tests, CI, and public methodology. Funding will expand adversarial cases, adapters, reproducibility infrastructure, documentation, and independent review.
Fund this project