Run a Side-by-Side Acceptance Test Against Your Own Answers

ROLE Act as an independent evaluator of an expert digital twin. CONTEXT Evaluate [GPT RESPONSES] against [MY REFERENCE RESPONSES] for the questions in [TEST QUESTIONS]. Apply [ACCEPTANCE PRIORITIES]. OBJECTIVE Determine where the GPT is ready to assist and where it still requires revision or human review. INSTRUCTIONS Create separate criteria for factual grounding, reasoning approach, voice alignment, usefulness, uncertainty handling, and authority boundaries. Treat my reference responses as evidence of preferences, not automatic proof that every factual claim is correct. Flag questionable claims in either set for verification. Compare responses question by question using observable evidence. Identify whether each mismatch likely comes from missing knowledge, unclear instructions, retrieval failure, or unrealistic expectations. Recommend the smallest targeted fix and a follow-up test for each important issue. Separate blocking defects from improvements that can wait. Do not reward superficial phrase matching when the reasoning or boundaries are wrong. CONSTRAINTS Avoid fabricated statistics and flag assumptions. Scores are rubric-based judgments, not scientific measures of how completely the GPT has cloned me. OUTPUT FORMAT Provide the rubric, comparison table, defect backlog, targeted fixes, and a launch recommendation with explicit conditions.