Before and after the eval loop

Reveal what changes when Nora turns prompt tweaking into an eval-backed refinement loop. Before: six prompt versions, no passing standard, no stable test. After: one success criterion, one patch, one retest. The failure changed from a feeling to a criterion: missing cited risks and excessive length. That makes the next prompt patch measurable. The patch changed one variable at a time. A source citation rule is tested separately from the length constraint, so Nora learns what each change did. The same input is reused for retest. That controls the comparison and prevents a new example from hiding whether the prompt…

Sign up free — one personalized lesson every day, matched to your role and goals.

Already have an account? Sign in