Catch the bug

Plausible-but-wrong code where the naive tests pass. Name the defect under each card's target time — this is what the AI-enabled round actually tests: spotting wrong AI output, not prompt craft.

0/23 done