Data Uncertainty in Football

Uncertainty is generated from three primary sources, each contributing to the overall data quality assessment.

• Data completeness: How many signals have the data they need. Missing injury reports, unavailable xG data, or absent fixture information each raise specific uncertainty flags.
• Data freshness: How recently the data was updated. Injury reports older than 48 hours, form data from early season, or stale odds data all trigger freshness warnings.
• Signal conflicts: When signals disagree — for example, strong recent form but weak underlying xG metrics — this creates directional uncertainty. The conflicting signals may both be correct (unsustainable results) or one may be noise.
• Sample size: Early season, newly promoted teams, and teams with unusual schedules often have insufficient match data for reliable analysis. Each signal declares its own minimum data requirements.

Uncertainty is always calculated and always reliable — it is the one signal that cannot itself be uncertain.

• Most useful early season: In the first 5-8 matchdays, uncertainty is typically elevated across most signals. This is the period where the signal adds the most value by preventing overconfidence in thin data.
• Important for newly promoted teams: Teams entering a new division lack historical data at that level. Uncertainty flags this gap clearly.
• Reduces naturally as season progresses: By mid-season, most signals have sufficient data and freshness. Uncertainty levels typically decrease to their floor values.
• Always check alongside other signals: High uncertainty in one area (e.g., availability) may be offset by low uncertainty in another (e.g., league position). The overall assessment weights each signal by its own confidence level.