Accuracy

How accurate is the SigningLab model.

The market gets roughly one signing in three right; that base rate, and the club rankings behind it, are published on the facts page under their own definition. This page reports something different: the SigningLab model’s own classification accuracy, measured only on the signings its models evaluate, under a published binary rule, recorded before outcomes are known. The two figures use different populations and different rulers and should not be read as one comparison. We publish the result every season, favorable or not; the full verdict-by-outcome matrix, baselines and definitions are on the verification page. What stays proprietary is the model’s internal architecture, its features and their weighting, not the results.

~81%
Model classification accuracy, 2025 average, evaluated subset
1.85×
Strict-success lift of recommended signings, 2024–2025
21
Leagues tracked
Ex-ante
Recorded before the outcome

Model accuracy by league (2025)

Each row reports the validated result of the SigningLab Model in that league over the published period, on the signings its models evaluate. Market conversion rates by league are published separately on the facts page; they use a different ruler and population and are not shown side by side here.

League20242025SigningLab Model, period
Saudi Pro League90%88%89%
Primera A (Colombia)90%86%88%
Liga MX (Mexico)86%84%85%
Liga 1 (Peru)85%85%85%
Super Lig (Turkey)83%86%84%
Primera Division (Chile)85%83%84%
Brasileirao Serie B84%84%84%
Serie A (Italy)86%81%83%
Liga Profesional (Argentina)85%81%83%
Bundesliga81%81%81%
Premier League82%80%81%
Brasileirao Serie A80%81%80%
J1 League (Japan)79%81%80%
La Liga (Spain)79%81%80%
Eredivisie (Netherlands)82%78%80%
Ligue 1 (France)80%79%80%
Primeira Liga (Portugal)76%82%79%
MLS (USA)78%78%78%
Championship (England, 2nd tier)75%80%78%
Pro League (Belgium)78%76%77%
Premiership (Scotland)74%75%74%

Percentages rounded. The model figure is the validated, out-of-sample result of the SigningLab Model in that league and period, on the subset of signings its models evaluate. The recommendation is always binary; the later outcome has three classes, and neutral is a tolerance band compatible with either signal. No model is correct every time; the commitment is reduced uncertainty and a record that can be audited.

How signing success is defined

A signing is counted as a success when the player performed at or above the baseline expected for the level of investment, read on minutes, performance, availability and collective contribution. Outcomes at the level expected count as neutral; clear underperformance counts as a miss. The same definition is applied across leagues and seasons so the comparison is consistent.

How the model is validated

Every projection is recorded and time-stamped before the outcome is known, so there is no future data and no hindsight. Validation is out-of-sample and walk-forward: the model is trained only on seasons before the one it is scored on, and each competition is calibrated on its own terms. The record is published season after season, whether the result is favorable or not.

The four figures, kept distinct

To read the numbers correctly, four ideas should not be confused with one another.

Market hit rate. The share of all signings in a league that worked, with no model involved. Roughly one in three across the leagues and seasons we track.
Club hit rate. The same measure for a single club, used in the public rankings to compare how well clubs convert signings into performance.
PIS, the public index. A retrospective, public evaluation index used in the annual reports. It is a simplified, transparent version meant for general understanding, not the commercial model.
Model accuracy. The validated, ex-ante hit rate of the SigningLab predictive model, the figures on this page. The commercial models are distinct from the public index and are not disclosed in their internals.

Last updated: 2026. Source: SigningLab annual validation. Machine-readable league data at facts.html and rankings.json.