Everything the report measures, on one page. Pick skill is the stated chance against what happened: Brier 0.25 is a coin flip, so Brier skill is how much better than a coin flip the stated chances were. Paper units are 1u per pick at the model's own fair price, never money. Written 2026-11-12 09:53 UTC.
| Sport | Graded | W-L | Won | Brier | Brier skill | Calibration slope | Paper units | 60%+ |
|---|---|---|---|---|---|---|---|---|
| tennis | 2033 12 pending | 1102-931 | 54.2% stated 54.6% | 0.2410 | 3.6% | 1.23 | -18.48u | 272-109 · +21.61u |
| soccer | 1694 57 pending | 845-453-396 | 49.9% stated 48.3% | 0.2368 | 5.3% | 1.47 | +38.18u | 137-18-27 · +23.95u |
| baseball | 339 10 pending | 199-140 | 58.7% stated 57.0% | 0.2409 | 3.6% | 0.87 | +10.71u | 57-33 · -0.35u |
| football | 46 14 pending | 30-16 | 65.2% stated 69.6% | 0.2067 | 17.3% | — | -3.33u | 24-10 · -2.35u |
| lol | 31 1 pending | 16-15 | 51.6% stated 62.8% | 0.2306 | 7.8% | — | -6.42u | 10-4 · -0.44u |
| Tennis model | Held-out acc (seasoned) | Brier | Warming-up acc | Brier |
|---|---|---|---|---|
| duel-elo-v1 baseline (the lab's v1 replay reproduces predict-41's v1 backtest; live now: duel-elo-v4) | 59.4% | 0.2352 | 55.2% | 0.2414 |
| Lab's best rule: History + pre-season seed for newcomers + margin-weighted (z -3.83; warming-up z -6.99) | 62.7% | 0.2263 | 64.2% | 0.2177 |
Shadow v2 forward test: 167 locked, 151 graded, 12 pending · v1 Brier 0.2399 vs v2 0.2317 · live duel-elo-v2/duel-elo-v3/duel-elo-v4 0.2136 vs v1 0.2373 on 103.
151 tennis picks carry a Polymarket price stamped once before the start (vig removed; read-only, never an order). On the 139 graded so far: model Brier 0.2157 (won 66.9%) against the market's 0.2114 (won 62.6%). They backed the same player 105 times; of the 34 splits the model was right 20. The research bar: the betting market is usually the hardest benchmark to beat.
| Manager | Bankroll | Settled P&L | ROI | W-L | Worst drop | Open · at stake |
|---|---|---|---|---|---|---|
| Guardian risk management: agreed favourites only, small stakes, tight exposure | 98.07u | -1.93u | -7.3% | 20-13 | 3.3% | 3 · 2.26u |
| Hunter high risk, high return: longer prices, bigger stakes, the underpriced underdog | 96.05u | -3.95u | -7.7% | 9-15 | 19.1% | 4 · 10.19u |
Picks: 0 in play now · 590 finals captured all time (every conviction) · 412 with their plays in the learner capture · 340 score timelines archived. Every listed match: 104 tracked (tennis 102, football 2), 0 in play · 595 finals in the all-time archive.
| Day | Finals archived (every listed match) | Picked matches captured with plays | Market prices stamped | Shadow picks graded |
|---|---|---|---|---|
| 2026-11-12 | 20 | 15 | 7 | 15 |
| 2026-11-11 | 2 | 0 | 3 | 3 |
| 2026-11-10 | 22 | 15 | 8 | 13 |
| 2026-11-09 | 28 | 23 | 9 | 16 |
| 2026-11-08 | 82 | 73 | 20 | 13 |
| 2026-11-07 | 91 | 79 | 26 | 19 |
| 2026-11-06 | 48 | 33 | 10 | 16 |
2026-11-09: PASS · 10 judged · 10 settled