PROMPT-DEBUGGING5 MIN READ
Sort weak outputs by root cause
Sort weak AI outputs into instruction, evidence, and evaluation root-cause buckets.
Sort the weak-output scenes by the root cause you would debug first. Instruction failure Evidence failure Evaluation failure Output ignored the requested schema instruction Answer invented a refund policy evidence Answer rambled instead of briefing Looked fine until QA rejected it evaluation Model missed facts never provided prompt-debugging Moved from symptom labels to root-cause buckets. Instruction, evidence, evaluation: pick one before patching.
Read the full lesson
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in