local-benchmark-runner-public

Contamination-resistant LLM evaluation harness: sealed benchmark banks with fabricated evidence, hashes published before measurement, verified task completion, and family-level statistics. Works with any OpenAI-compatible endpoint.