Why this needs reviewThe response is not strongly supported by retrieved evidence (best hybrid 0.48; strong threshold 0.52).
- Best grounding score: 0.48
- Sentence match: 0.42
- Keyword overlap: 0.62
Full scorer narrative
The response is not strongly supported by retrieved evidence (best hybrid 0.48; strong threshold 0.52).
Next steps
- Inspect retrieved docs below — scores blend best-sentence similarity with keyword overlap.
- Tighten retrieval or add citations in the agent prompt.
What we measured
- Hybrid grounding (per chunk): best 0.48 (strong ≥ 0.52)
- Mean hybrid across chunks: 0.48
- Best raw sentence match: 0.42 (short answers often score low vs. whole paragraphs)
- Lexical overlap with sources: 0.62
- Blend: standard blend (70% best-sentence + 30% keyword)
- Weak / strong cutoffs: 0.35 / 0.52
EvidencePrompt & response