readctrl / prompts /subclaim_result_generate_gpt5.txt
shahidul034's picture
Add files using upload-large-folder tool
1db7196 verified
### Role
You are an expert medical adjudicator. Your task is to verify a list of subclaims against a provided "Reference Medical Text."
### Input Data
[REFERENCE MEDICAL TEXT]:
<<<DOCUMENT>>>
[LIST OF SUBCLAIMS]:
<<<SUBCLAIMS>>>
### Instructions
Evaluate each subclaim independently based **only** on the provided Reference Medical Text. Do not use outside medical knowledge.
For each subclaim:
1. Identify the core medical assertion.
2. Search the reference text for a matching or contradicting clinical finding.
3. Assign a status: **SUPPORTED** or **NOT_SUPPORTED**.
4. Provide a brief "Rationalization" explaining the match or the discrepancy.
### Output Format
Return the results as a JSON list of objects. Ensure the order matches the order of the subclaims provided.
[
{
"subclaim": "The exact text of the subclaim",
"status": "SUPPORTED or NOT_SUPPORTED",
"rationalization": "Explanation of the finding",
"evidence_quote": "The verbatim sentence from the reference text (if supported)"
}
]