Turn the core rule into observable checks
Write checks that describe something a tester can see. For the charge-and-launch example, verify that holding longer changes the charge meter, releasing once launches one projectile, landing inside the gate increases the score, and landing outside it produces a clear miss. Include one opposite case for each success condition. If both an inside and outside landing score, the rule is not behaving as described.
Separate rule failures from presentation issues
Play the same short scenario while watching one thing at a time. First ignore artwork and check input, scoring, collision, and restart behavior. Next ask whether the meter, gate, and feedback communicate those rules. A missing score animation is a presentation problem; a score awarded for a miss is a logic problem. Logging these separately helps avoid rewriting a working mechanic to fix a confusing label.
Questions about AI game testing
How many cases should an early AI game test include?
Begin with a small set tied to one mechanic: one ordinary success, one near miss, one clear failure, and one restart. Add cases only when a new rule creates a new risk.
Can browser automation judge whether a prototype is fun?
Automation can check outcomes such as score changes, button response, and navigation. A person still needs to judge whether feedback is understandable and the challenge feels fair.