Graded against results
Every fight projection we publish gets graded here after the card — no deletions, no do-overs. The record starts July 11, 2026 (UFC 329), the first card we projected live.
Running record: 10-4 (71% of picks correct across 1 card). For context, always picking the betting favorite historically lands ~65% — beating that consistently is the bar we hold ourselves to before anything here becomes more than research.
UFC 329: McGregor vs. Holloway 2
July 11, 2026 · model 10-4
| Our pick | Our % | Result | |
|---|---|---|---|
| ✓ | Max Holloway | 70% | Max Holloway won — KO/TKO R1 |
| ✗ | Benoît Saint Denis | 59% | Paddy Pimblett won — Submission R1 |
| ✓ | Mario Bautista | 56% | Mario Bautista won — Decision - Unanimous R3 |
| ✗ | Lone'er Kavanagh | 68% | Brandon Royval won — Submission R3 |
| ✗ | Terrance McKinney | 64% | King Green won — KO/TKO R1 |
| ✓ | Robert Whittaker | 54% | Robert Whittaker won — KO/TKO R3 |
| ✓ | Gable Steveson | 67% | Gable Steveson won — KO/TKO R1 |
| ✓ | Adrian Yanez | 55% | Adrian Yanez won — KO/TKO R1 |
| ✓ | Luke Riley | 67% | Luke Riley won — KO/TKO R1 |
| ✗ | Tracy Cortez | 51% | Wang Cong won — Decision - Unanimous R3 |
| ✓ | Damian Pinas | 73% | Damian Pinas won — KO/TKO R1 |
| ✓ | Farid Basharat | 80% | Farid Basharat won — Decision - Unanimous R3 |
| ✓ | Ryan Gandra | 54% | Ryan Gandra won — KO/TKO R1 |
| ✓ | Alessandro Costa | 63% | Alessandro Costa won — Submission R2 |
Backtest: how the model grades on history
Out-of-sample walk-forward — for each year the model trains only on earlier years, then picks every fight. These picks were computed retrospectively, never published; they're also shown fight-by-fight on each past event page.
Backtest record: 2511-1588 (61.3% across 4,099 fights, 2019–2026). Honesty first: on the 2,718 fights with closing lines, the betting market is the sharper forecaster (log loss 0.601 vs our 0.653) — favorites hit 68%. That's exactly why leans are research, not picks.
| Year | Picks | Record | Hit % | Log loss |
|---|---|---|---|---|
| 2019 | 553 | 323-230 | 58.4% | 0.6642 |
| 2020 | 492 | 315-177 | 64.0% | 0.6426 |
| 2021 | 549 | 329-220 | 59.9% | 0.6648 |
| 2022 | 558 | 338-220 | 60.6% | 0.6553 |
| 2023 | 556 | 348-208 | 62.6% | 0.6440 |
| 2024 | 562 | 342-220 | 60.9% | 0.6576 |
| 2025 | 563 | 353-210 | 62.7% | 0.6487 |
| 2026 | 266 | 163-103 | 61.3% | 0.6486 |
| When we say… | Fights | It happens |
|---|---|---|
| 55% | 1167 | 58% |
| 64% | 826 | 70% |
| 74% | 317 | 76% |
| 83% | 73 | 88% |
A pick = whichever fighter our model gives more than 50%. Picks on small-sample fights count here too — the record reflects everything we publish. Research, not betting advice — 21+.