Team Value · participant
Arena shots model
The Arena’s own reference model. It delivers through the same public API as every other participant, on the same deadlines, and is scored by the same weekly run — it is labelled as ours wherever it appears, and it is not the baseline.
The Arena's own reference model: shots for, per team per match.
Weekly scores
Newest week 2026-W43
| Week | Skill score | Interval | Somers’ D | Fixtures scored |
|---|---|---|---|---|
| 2026-W43 | -0.013 | -0.11 to +0.08 | 0.252 | 202 |
| 2026-W42 | -0.049 | -0.16 to +0.05 | 0.237 | 153 |
| 2026-W41 | -0.079 | -0.20 to +0.05 | 0.213 | 106 |
| 2026-W40 | -0.079 | -0.19 to +0.04 | 0.213 | 106 |
| 2026-W39 | -0.079 | -0.21 to +0.04 | 0.213 | 106 |
| 2026-W38 | -0.196 | -0.37 to -0.01 | 0.149 | 50 |
The counting
Every published figure, from its pair counts
Horizon h0 builds each signal from every match before the fixture; h1 withholds the most recent round and h2 the two most recent. The skill score is the participant’s Somers’ D minus the baseline’s, averaged over the three. D leaves drawn pairs out of its denominator, so a perfect ordering reaches 1. τ-b keeps them in both halves of its ratio and therefore cannot: its ceiling, √(1 − n_Y/n₀), is the most it could have been on these results, and is what makes the two scales comparable.
| Horizon | Somers’ D | τ-b | Ceiling | n₀ | C | D | n_S | n_Y |
|---|---|---|---|---|---|---|---|---|
| h0 | 0.2600 | 0.2381 | 0.914 | 20,301 | 10,676 | 6,262 | 50 | 3,325 |
| h1 | 0.2466 | 0.2260 | 0.914 | 20,301 | 10,544 | 6,357 | 89 | 3,325 |
| h2 | 0.2508 | 0.2301 | 0.914 | 20,301 | 10,563 | 6,306 | 138 | 3,325 |
| Horizon | Somers’ D | τ-b | Ceiling | n₀ | C | D | n_S | n_Y |
|---|---|---|---|---|---|---|---|---|
| h0 | 0.2685 | 0.2476 | 0.914 | 20,301 | 10,626 | 6,068 | 338 | 3,325 |
| h1 | 0.2861 | 0.2642 | 0.914 | 20,301 | 10,758 | 5,902 | 402 | 3,325 |
| h2 | 0.2425 | 0.2252 | 0.914 | 20,301 | 10,295 | 6,179 | 615 | 3,325 |
The interval on each week is paired and resampled 1,000 times over the fixtures of that week’s season. Participant and baseline are resampled together, which is what resolves the difference between them far better than either number alone. Read the protocol · the fixtures these counts are over · back to the board.