We set up a simple method: every piece of feedback must be classified (bug, readability, rhythm, balance, UX) and accompanied by context (mode, situation, player objective). No "like/dislike" without examples.
This allows iterating without scattering: we know exactly what we're changing, why, and what impact we expect.
What we ask testers:
- "What were you trying to do?"
- "What prevented you?"
- "At what point did you understand (or not)?"