evals/README.md

# Regression and Evaluation Suite

This package contains 130 behavioral cases across source provenance, definitions, accounting, valuation, forensics, risk, management, forecast errors, confidence, EVoI, sector routing, red team, and workflow routing. The standard-library tests validate contract integrity and deterministic functions. Model behavior can be scored by recording observed tags against each case's must and must_not assertions; no network or server is required.