SAGE-QA Framework Distinguishes LLM Paraphrasing from Hallucinations
November 13, 2026
SAGE-QA uses Semantic Equivalence Scoring and Re-Validation Scoring to evaluate closed-context question answering on SQuAD 2.0. This framework addresses the limitation of Exact Match metrics by separating valid lexical paraphrasing from actual factual errors.
HOW THIS AFFECTS YOU
●
builderYou can stop penalizing your models for using different words to say the same thing.
●
researcherYou can use this to get a more accurate view of how well your models actually understand context.