bootstrap-ci

OMK — Evidence-backed evaluation and observability for prompts, RAG, skills, agents, and workflows. Native Codex, Claude Code, and DeepSeek Harness support.

Do agents make for good offensive & defensive coordinators in football? This is an adversarial-agent arena for short red-zone strategy contests. OC & DC agents compete through simultaneous legal play calls, scarce tactical resources, partial observability, in-game belief updates, and seeded reproducibility.

Agent evaluation harness: golden set regression + paired bootstrap CI + 16 scorers + 30 cases across 6 failure modes. v0.2.0 baseline: deterministic 30/30, flaky 5/30, 5 scorers regressed at 95% CI.

Diversity Indices with Statistical Inference

Statistically rigorous RAG evaluation framework featuring empirical 95% bootstrap confidence intervals, paired permutation hypothesis testing, and zero-API offline benchmarking.