Claude Auditable Reasoning Without Raw Thinking Output
Production systems should not depend on exposing hidden chain-of-thought to users, logs, or downstream systems. Instead, ask Claude for structured, auditable outputs that document evidence and missing context.
Analyze the clinical summary in <clinical_notes>{{NOTES}}</clinical_notes> against <guidelines>{{GUIDELINES}}</guidelines>. Do not expose internal chain-of-thought. Instead, return auditable outputs in three sections:
1. <supporting_quotes>: Verbatim passages from the notes and guidelines.
2. <evidence_gaps>: Any missing data points where evidence is insufficient to answer.
3. <synthesis>: Your grounded summary.Fill in the blanks (2)
Nobody has confirmed this prompt yet. If you try it, tell others whether it worked.
No comments yet.