Prompt debugging diagnostic deck
Recall a concise diagnostic checklist for weak AI outputs.
FCPR What is the four-step micro-loop for a weak AI output? Failure, cause, patch, retest Symptom label vs. root-cause label Root-cause labels point to the next prompt change. Someone says: "Just make the prompt better." Use when a teammate wants a fast rewrite without diagnosis. Your line Let's name the failure first: what did the output do, and what would a passing answer have done? Rewriting immediately and losing the failure evidence. It shifts the conversation from taste to observable criteria. What are the five common cause buckets? Instruction, context, evidence, evaluation, capability. Pick the primary bucket before choosing a…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in