{
"model": "jev-latest",
"state": {
"query_excerpt": "{{query-excerpt}}",
"candidate_passage": "{{candidate-passage}}"
},
"questions": {
"is_the_source": {
"type": "noul",
"instructions": "Does the candidate passage supply the specific {{thing-the-query-needs}} that the query excerpt relies on?",
"criteria": {
"true": {"what": "The candidate states or establishes the specific rule, fact, figure, or step the query excerpt depends on"},
"false": {"what": "The candidate is merely on a similar topic; it does not supply the specific thing the query excerpt needs"}
}
}
}
}
Usage: take the top 20 to 40 candidates from BM25 or vector search, fire one request per query-candidate pair from a thread pool, then sort the shortlist by is_the_source descending. The false criterion is the important half of the question. Without it the moSemantic Rerank of a Retrieval Shortlist
Generate a judgment on whether a candidate passage provides specific information needed by a query excerpt, producing a JSON output.


