Your pass output looks 95% similar to the original document. Sounds great, right. Except that missing 5% was liability clause in a legal contract.MTF from Lama index content faithfulness is arguably the most critical dimension in pass bench. The first document benchmark for AI agent. If an agent's context is incomplete or fabricated, every downstream decision is compromised.Our metric for this is the content faithfulness score. Traditional metrics like blue use text similarity. They are great at catching ty ...