Start with a baseline play session
Before revising, describe the core experience and capture what already works. For a falling-block concept, you might note that players understand the left-right movement, but do not notice the previewed next shape. Keep the input, goal, and first obstacle constant in your review notes. This baseline is a reference for the next pass, not a demand that every detail stay fixed forever.
Revise one instruction and repeat the test
Write a narrow change request that names the behavior and a way to recognize success. For the shape preview, ask that the next piece appear in a labeled preview box before the current piece locks; then check whether a player can identify it during a short run. Do not simultaneously add power-ups, a new art style, and a level system. Those additions create more variables and make it harder to tell whether the specific revision helped.
Questions about AI game iteration workflow
How many changes should I make in one AI game iteration?
Make one major change at a time when you need to know what caused the result. Small supporting edits can be grouped if they all address the same observed problem.
What should I do when a revision fixes one issue but creates another?
Record both outcomes, then test a targeted adjustment such as changing cue placement. Preserve the original case in your regression checks so it does not quietly return.