Ranking methodology
GamersPilot ranks a Steam backlog with fixed math, not AI inference.
Every candidate receives normalized signals, each signal is multiplied by the visible preset or user-selected weight, and the contributions are added. The same library, settings, filters, and timestamp produce the same order.
Signals in the score
- Rating
- The average of available critic and personal ratings, normalized to a 0–1 range. Missing ratings use a neutral midpoint, never a fabricated score.
- Unplayed
- A binary signal for games with zero recorded playtime or an explicit unplayed status.
- Neglect
- Time since last play, rising gradually across one year. A never-opened game receives the maximum neglect signal.
- Session fit
- How closely the estimated session length matches the time window selected by the player.
- Completion momentum
- Known completion percentage when available; otherwise a transparent fallback based on playing status and recorded playtime.
- Popularity
- A normalized popularity signal. Missing data again receives a neutral midpoint.
- Free-title confidence
- An optional small bonus only when the source says a title is free with medium or high confidence.
Penalty and tie-break order
An abandoned game receives a fixed 0.75 penalty. Exact score ties are then resolved by critic score, lower total playtime, older last-played date, normalized title, and finally the internal game ID. There is no random shuffle hidden in the ranking.
What the model does not use
It does not read messages, friends, purchases, browsing history, or launcher passwords. It does not train on a player profile or call a language model. Inputs come from the public Steam library plus settings the player controls.
Verify it in the product
Each ranked result includes explanation chips for its strongest contributions. Open the planner to compare presets or tune the weights yourself.