ODDZY SPORTS INTELLIGENCE
VI
CALIBRATED MODEL STATUS

Research laboratory

Oddzy separates data quality, historical model reliability and matchup uncertainty. A fallback calculation can never be published as a production forecast.

Historical results49,520Versioned artifact loaded
Backtested fixtures4,639Chronological out-of-sample evaluation
Brier score0.181Lower is better; probability accuracy metric
Calibration error0.030Difference between stated confidence and observed frequency
Data qualityMEDIUM
80/100

Measures source coverage, freshness and matchup-specific sample depth.

Model reliabilityMEDIUM
72/100

Based on chronological out-of-sample probability performance.

Forecast certaintyLOW
48/100

The probability distribution has a clear leading outcome.

QUALITY GATE

historical-international-1.0.0

  1. Fixture identity, kickoff and venue are verified independently of the forecast.
  2. A versioned historical artifact must be present before any probability is published.
  3. A chronological pre-match backtest must be loaded; random train/test splits are not accepted.
  4. Model reliability comes from Brier score, calibration error, top-outcome accuracy and sample depth.
  5. Forecast certainty measures how decisive the current outcome distribution is. Close matches remain low-certainty even when the model is reliable.
  6. Confirmed lineups, injuries and expected minutes are still required before data quality can reach the highest tier.

Sources: martj42/international_results (CC0) · StatsBomb Open Data (attribution required). Event coverage: StatsBomb Open Data: World Cup 2018/2022 and UEFA Euro 2020/2024.