Quality criteria
Define what a useful and correct result looks like.
Evaluation
Measure usefulness, not just output
A convincing AI project needs task-specific checks, failure analysis, and human review.
Failure cases
Test ambiguity, outdated data, unsupported claims, and unsafe actions.
User experience
Evaluate whether the workflow saves time without reducing control.
Evaluation table or test resultsReplace with an optimized WebP using the included media script.