fix: skip adversarial regression gracefully when API keys not configured

feat: physician adjudication integration (lightweight pointer to SG2)

fix: update stale OpenEM condition count 363 → 370

docs: surface campaign engine, methodology, governance, and attack taxonomy in README

Wire defer differentials RAG into defer runner (#1)

ci: add CODEOWNERS for workflow review trail

ci: add permissions block to adversarial-regression workflow

docs: update exploit_families and CLAUDE.md with resolved judge agreement

fix: judge JSON parse retry + regrade corrects κ=0.137 → 90.6% agreement

EF-016: both-calibrated regrade reveals substantive judge disagreement (κ=0.137)

EF-016: MCI calibration resolves GPT judge bias (agreement 69%→72%)

EF-016: calibrated judge regrade (κ=0.126→0.344)

judge calibration: Level 2 vs 1 boundary fix resolves 3/3 disagreements

MSTS dual-judge validation: κ=0.400 on non-MCI conditions

EF-016: judge asymmetry validation + GPT preamble regression forensics