Define measurable success criteria before you write the prompt

"Good answers" cannot be tested. Before you write the prompt, decide what success looks like in terms you can measure.

Anthropic's eval guide recommends criteria that are specific and measurable, and notes that most apps need several at once:

• accuracy
• format
• tone
• latency
• cost
• safety

Write each criterion with a number and a test set. For example:

• "Category matches the human label on at least 90% of 300 real tickets."
• "No answer includes another customer's data."

Pick thresholds from what the feature needs. Then use them as the ship rule: a prompt or model change goes out only when it meets the criteria.