Libero's fair value model reads a season of form: shares of minutes and starts, league-adjusted per-90s, team points and goal difference. Five games in, those numbers are extreme (a full share of minutes is common, team goal difference is spread wide) and raw fair values for regulars come out 43% to 55% above market. The fix is to blend each form feature with the player's and team's previous season.

To choose the weight, 25-26 was rebuilt from only the matches played by 15 September, 13 October, 15 December and 15 February, and scored by a model trained on the data as it stood before the season began. Each variant is scored against the market value on that date.

Read as accuracy, Figure 1 says to ignore early-season form entirely. Read as a description of the target, it says the market value on a date is itself mostly last season, with this season's form entering slowly enough that sixteen games are not yet enough to overtake it. A model scored against the market in autumn is rewarded for lagging.

The other reference is the model's own verdict on the complete season, measured as mean absolute log distance.

The May verdict is built from this season's form, so the raw variant converges on it partly by construction and last-season-only cannot. The informative part is the early dates, where the blend is closest to May while staying near the best market fit: 0.418 from May against 0.462 for last season alone, at R² 0.737 against 0.757. Libero uses that weight, 44% on this season after eight games and 71% after 24.