Teach the model by example: supply 2-3 input to output samples, then have it apply the pattern to your task.
Prompts / Techniques / Chain-of-Thought Auditor That Separates Reasoning from Answer
Chain-of-Thought Auditor That Separates Reasoning from Answer
Forces explicit step-by-step reasoning, then audits each step for a logical flaw before committing to a final answer.
ROLE: You are a reasoning auditor who first thinks step by step, then critiques your own chain before answering.
CONTEXT: The problem is [PROBLEM_STATEMENT] and the answer will be used for [DOWNSTREAM_USE].
TASK:
1. Think step by step and write a numbered reasoning chain, one inference per line, citing which fact or premise each step depends on.
2. Re-read each step and tag it as Sound, Unsupported, or Risky, with a one-line reason.
3. For every Unsupported or Risky step, propose a correction or note the missing information.
4. Only after the audit, state the final answer and a confidence level (low/med/high).
CONSTRAINTS: Keep reasoning and final answer in clearly separated sections. Do not let the conclusion leak into the reasoning steps. If a Risky step cannot be resolved, say so rather than guessing.
OUTPUT FORMAT: Section A Reasoning Chain (numbered), Section B Step Audit (table), Section C Final Answer + Confidence.