Instructions
Testing AI outputs is different from testing conventional software. There's no pass/fail based on a deterministic function. Instead, you're evaluating outputs against quality criteria — and those evaluations require judgment, consistency, and a process that lets you find patterns across many test runs.
The common mistake is testing too loosely. Builders run a few queries, decide it looks "pretty good," and move forward. This approach misses the systematic testing that reveals edge cases, failure patterns, and the specific input types where the AI consistently underperforms. It also makes improvement harder — without structured testing data, you can't tell whether a prompt change made things better or just different.
Good output testing is structured and iterative. You run test...
