Compact Proofs of Model Performance via Mechanistic Interpretability | Read Paper on Bytez