RKB109's picture
Publish artifacts for rag-evaluation-lab-20260720
9f198fa verified
Raw
History Blame Contribute Delete
4.76 kB
{
"format": "daily-project-prototype-v1",
"project": "rag-evaluation",
"title": "RAG Evaluation Lab",
"domain": "ai-evaluation",
"mode": "classifier",
"labels": [
"retrieval-miss",
"unsupported-answer",
"citation-gap",
"abstention-error"
],
"prototypes": {
"retrieval-miss": {
"the": 15,
"expected": 3,
"document": 3,
"never": 3,
"appeared": 3,
"in": 5,
"top": 3,
"results": 3,
"relevant": 6,
"evidence": 3,
"was": 3,
"absent": 3,
"from": 3,
"retrieved": 3,
"context": 3,
"an": 4,
"operations": 2,
"review": 2,
"for": 2,
"evaluation": 2,
"case": 2,
"ranking": 3,
"placed": 3,
"irrelevant": 3,
"policy": 3,
"above": 3,
"correct": 3,
"section": 3,
"item": 3,
"fell": 3,
"outside": 3,
"accepted": 3,
"retrieval": 3,
"window": 3
},
"unsupported-answer": {
"in": 4,
"an": 4,
"operations": 2,
"review": 2,
"the": 6,
"response": 4,
"makes": 2,
"a": 4,
"claim": 4,
"not": 2,
"present": 2,
"any": 2,
"source": 2,
"answer": 2,
"faithfulness": 2,
"failed": 2,
"despite": 2,
"available": 2,
"context": 2,
"for": 2,
"evaluation": 2,
"case": 2,
"invented": 2,
"product": 2,
"limit": 2,
"generated": 2,
"cannot": 2,
"be": 2,
"traced": 2,
"to": 2,
"retrieved": 2,
"evidence": 2
},
"citation-gap": {
"the": 2,
"answer": 4,
"is": 2,
"correct": 2,
"but": 2,
"has": 2,
"no": 2,
"source": 2,
"reference": 2,
"citation": 2,
"coverage": 2,
"failed": 2,
"for": 3,
"an": 3,
"otherwise": 2,
"supported": 2,
"evaluation": 1,
"case": 1
},
"abstention-error": {
"the": 4,
"system": 2,
"answered": 2,
"even": 2,
"though": 2,
"evidence": 2,
"was": 2,
"insufficient": 2,
"answer": 2,
"policy": 2,
"should": 2,
"have": 2,
"triggered": 2,
"abstention": 2,
"in": 1,
"an": 1,
"operations": 1,
"review": 1
}
},
"idf": {
"relevant": 1.847298,
"evidence": 1.847298,
"was": 2.252763,
"absent": 2.252763,
"from": 2.252763,
"retrieved": 1.847298,
"context": 1.847298,
"answer": 1.559616,
"faithfulness": 2.252763,
"failed": 1.847298,
"despite": 2.252763,
"available": 2.252763,
"citation": 2.252763,
"coverage": 2.252763,
"for": 2.252763,
"an": 2.252763,
"otherwise": 2.252763,
"supported": 2.252763,
"the": 1.559616,
"policy": 2.252763,
"should": 2.252763,
"have": 2.252763,
"triggered": 2.252763,
"abstention": 2.252763,
"item": 2.252763,
"fell": 2.252763,
"outside": 2.252763,
"accepted": 2.252763,
"retrieval": 2.252763,
"window": 2.252763,
"generated": 2.252763,
"claim": 2.252763,
"cannot": 2.252763,
"be": 2.252763,
"traced": 2.252763,
"to": 2.252763
},
"documents": [
{
"id": "eval-01",
"label": "retrieval-miss",
"text": "Relevant evidence was absent from retrieved context.",
"metadata": {
"synthetic": true,
"domain": "ai-evaluation"
}
},
{
"id": "eval-02",
"label": "unsupported-answer",
"text": "Answer faithfulness failed despite available context.",
"metadata": {
"synthetic": true,
"domain": "ai-evaluation"
}
},
{
"id": "eval-03",
"label": "citation-gap",
"text": "Citation coverage failed for an otherwise supported answer.",
"metadata": {
"synthetic": true,
"domain": "ai-evaluation"
}
},
{
"id": "eval-04",
"label": "abstention-error",
"text": "The answer policy should have triggered abstention.",
"metadata": {
"synthetic": true,
"domain": "ai-evaluation"
}
},
{
"id": "eval-05",
"label": "retrieval-miss",
"text": "The relevant item fell outside the accepted retrieval window.",
"metadata": {
"synthetic": true,
"domain": "ai-evaluation"
}
},
{
"id": "eval-06",
"label": "unsupported-answer",
"text": "The generated claim cannot be traced to retrieved evidence.",
"metadata": {
"synthetic": true,
"domain": "ai-evaluation"
}
}
],
"graph_edges": [],
"confidence_threshold": 0.18,
"trained_on_synthetic_data": true
}