APEX / Data quality
What the data supports
Coverage shows what this release can support. A current roster snapshot does not prove historical playing time, health, or position. Missing evidence remains unavailable.
Source snapshots
Release coverage counts
Review flags by category
Roster coverage by team
Coverage and interpretation
Test the rating before trusting it.
These are winner forecasts from the experimental results-only team rating. Its probability scale was fitted on 2023 and frozen for 2024 and 2025. This evaluates neither the correctness of SOS ordering nor the APEX composite, player grades or betting performance.
| Season / role | Model | Games | Winner accuracy | Brier ↓ | Log loss ↓ |
|---|---|---|---|---|---|
| 2023 · Training calibration | Results-only rating | 792 | 68.5% | 0.2001 | 0.5818 |
| 2023 · Training calibration | Simple Elo baseline | 792 | 65.5% | 0.2335 | 0.6598 |
| 2023 · Training calibration | 50/50 probabilities | 792 | 50.0% | 0.2500 | 0.6931 |
| 2024 · Held out | Results-only rating | 798 | 66.5% | 0.2102 | 0.6081 |
| 2024 · Held out | Simple Elo baseline | 798 | 61.9% | 0.2353 | 0.6635 |
| 2024 · Held out | 50/50 probabilities | 798 | 50.0% | 0.2500 | 0.6931 |
| 2025 · Held out | Results-only rating | 808 | 67.1% | 0.2027 | 0.5886 |
| 2025 · Held out | Simple Elo baseline | 808 | 63.4% | 0.2324 | 0.6576 |
| 2025 · Held out | 50/50 probabilities | 808 | 50.0% | 0.2500 | 0.6931 |
Lower Brier and log loss are better. Accuracy awards half credit to exactly 50% forecasts. Forecasts use earlier UTC dates; historical publication times remain unverified. The Elo comparison is a fixed simple baseline, and these figures do not establish superiority over advanced models.