Methodology
GameIn separates factual observations from derived metrics, deterministic signals and commercial hypotheses. Scores are 0–100 indices, not probabilities.
Statistical normalization
Count-like metrics should use log transforms before robust percentile normalization. Cohorts should be defined by release age, platform/genre and period when sample size is sufficient. Missing values remain unknown rather than becoming zero.
Public score formulas
| Code | Score | Launch formula |
|---|---|---|
| GI01 | Game Attention | 25% peak players + 20% Twitch + 20% visits + 15% review scale + 10% seller + 10% playing |
| GI02 | Game Momentum | 25% player momentum + 20% review velocity + 15% Twitch + 15% interest + 15% seller/wishlist + 10% sentiment |
| GI03 | Pre-release Heat | 35% wishlist + 25% want-to-play + 15% visits + 15% Twitch + 10% release proximity |
| GI05 | Studio Momentum | 30% portfolio + 20% release pipeline + 15% hiring + 15% events + 10% financial + 10% reach |
| GI07 | Genre Momentum | 45% weighted games + 20% release growth + 15% pre-release heat + 10% studio growth + 10% attention growth |
Confidence
Cluster foresight
Forward-looking cluster outputs must retain a separate confidence score and, when historical cohorts are used, expose the comparable sample size and observed historical transition frequency. That frequency is evidence about history, not a guaranteed probability for the current city.
Versioning
Every public ranking or score artifact should carry metric_version or methodology_version, generated_at, source coverage and confidence. Same snapshot + same version must reproduce the same score.