What this page can support
A model description should name the target, input window, weighting, validation period and treatment of draws, extra time and penalties.
Ratings, probabilities and calibration are different
A rating orders or separates teams using defined inputs. A probability maps that difference to a specific event, such as a win in regulation. Calibration then checks whether events assigned about the same probability occur at about that rate in unseen matches.
Each step needs its own data, version and test period. A precise-looking percentage is not evidence that those steps were retained or validated.
Evidence checklist
- Define exactly what the probability predicts.
- Use out-of-sample matches for validation.
- Check whether stated probabilities are calibrated.
- Publish uncertainty and known blind spots.
- Label worked examples as illustrative.
Limit of the archive
No numeric example on this page is represented as a retained 26cup tournament run. The archive therefore explains what evidence a reproducible model would need instead of reconstructing missing inputs after the result.
Verify the tournament outcome
Use FIFA's official fixtures, results and final standings when you need the authoritative match record.