F10 Retrieval miss (the answer was not in the retrieved text)
RAG: the retrieved passages did not hold the answer, so the model could only guess or decline. The retriever is at fault, not the model.
How counterfix finds it
Free with gold labels: none of the gold passages was retrieved. Without labels, a judge decides whether the passages hold what the question needs (context_lacks_answer). F10 takes precedence over F4: the unsupported claim is the symptom.
Fixes it tries
In this order (a model profile can reorder them). Each one is kept only if it passes 2 of 3 re-runs.
| Intervention | Change |
|---|---|
gold_passage |
With the right passage in the context the answer passes: fix the retriever (more passages, a rewritten query, hybrid search) |
When no fix verifies
- Retrieve more passages (larger k)
- Rewrite or expand the query before retrieval
- Add keyword (hybrid) search or a reranker
- Re-index with smaller or overlapping chunks