How to Test an AI-Generated Game
A lightweight QA checklist for prompts, controls, edge cases and replayability.
Test generated games as untrusted software: verify every rule, reset path, device size, asset and external call rather than assuming a successful preview means the game is finished.
Start with a rule sheet
List each input, state transition, score event, win condition and loss condition. For every item, write one normal case and one edge case. This exposes vague design and gives you repeatable checks.
High-value tests
- Rapidly repeat every input and try simultaneous inputs.
- Pause, resize, rotate or background the app.
- Lose and replay ten times without refreshing.
- Test zero, maximum and negative-like boundary states.
- Run on the weakest target device and a slow connection.
- Inspect browser console and network requests where available.
Regression discipline
After each AI edit, retest the changed feature plus score, restart and save state. Keep versions so you can revert. A prompt that fixes one bug can silently break another system.
Test states, not only minutes of play
Generated games often appear correct during a normal run but fail at transitions. Create a state checklist and deliberately force every start, pause, win, loss and replay path. Repeat each path at least twice because stale variables often appear only after replay.
A compact test matrix
- Controls: press, hold, drag and release at screen edges.
- Timing: let timers expire and trigger events at the same moment.
- Scoring: verify minimum, maximum and repeated collisions.
- Replay: confirm all objects, audio, counters and effects reset.
- Devices: test the public build on the target screen and input method.
After losing once and pressing Replay, the timer resets to 60 but the enemy speed remains at the final difficulty. Expected: enemy speed returns to 120 pixels per second. Keep scoring and spawn timing unchanged. Acceptance test: lose and replay three times; each run must begin at 120.
Save a working version before each risky revision. When a test fails, change one system and rerun the smallest relevant group of checks before moving on.
Prioritize failures by player impact
Classify issues as blockers, confusing behavior or polish. A crash, impossible win state or broken replay blocks release. Unclear feedback and inconsistent controls damage understanding. Minor alignment and effects can wait. This prevents cosmetic fixes from hiding a broken loop.
After every AI revision, rerun tests for the system that changed and one nearby system it could affect. Before publishing, run the entire short regression list on a fresh session and the public build. Keep the list small enough that you will actually repeat it.
This page is a practical framework rather than a product-spec sheet. Validate platform-specific details before publishing a commercial project.