Run a clean prompt diff
Compare two prompt versions using a stable input, passing criteria, and targeted patch hypothesis.
Prompt A and Prompt B both generate customer-story drafts, but the team is choosing by feel instead of by the failure the patch was meant to fix. A clean prompt diff keeps input and criteria stable, changes one prompt element, and scores both outputs against the same passing standard. Choosing the more fluent answer can hide regressions. Prompt B may read better while dropping required evidence, length, or audience fit. Step 1 Freeze the test input: one customer interview excerpt, one target audience, one intended deliverable. If the input changes, you cannot tell whether the prompt patch or the new…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in