UFC track record
Every UFC prediction is logged before the fight and graded afterwards. Nothing is deleted. NFL track record
Live picks
The live UFC record starts with the first card after launch. Every pick is logged before the fight and graded after it, wins and losses alike. A draw or no contest counts as neither.
Backtest, 2016 to 2026
Each year was predicted using only fights before it. The market is the closing line: BestFightOdds' closing range via the UFC Data dataset (N. Bair, CC BY 4.0), with the sportsbooks' margin removed; ESPN's DraftKings line for the newest cards. Our ratings come from every UFC fight since 1994 (ufcstats.com).
| Winners picked | Log loss | |
|---|---|---|
| Closing betting line, 2020 on (3,450 fights) | 68.2% | 0.5990 |
| AZMango (what the pages show) | 68.5% | 0.5968 |
| Our ratings alone, same fights | 63.4% | 0.6436 |
| Closing line, 2016 on (5,348 fights) | 67.0% | 0.6053 |
| AZMango, 2016 on | 67.4% | 0.6026 |
The honest summary: we match the closing betting line; we do not beat it in any way you could bet on. Blending our ratings into the closing line lowered the log loss over the last 1,000 fights (0.5857 against 0.5863), so the guard rail has the blend on. That gap is tiny, far too small to bet on, and the guard rail switches the blend off the week it stops helping. What we add is the explanation, how and when a fight is likely to end, a number for fights no book has priced yet, and probabilities you can check.
How it ends (who wins and how: six outcomes), 5,356 fights: our method model's log loss 1.556, against 1.632 for each division's usual KO/submission/decision shares. Against the sportsbooks' method-of-victory prices (784 fights from 2024 on, the ones ESPN kept): books 1.489, our model 1.521, our win chance with the books' split 1.482. The books are better at how a fight ends, so where they price it the pages use their split.
Total rounds, 2,849 fights with a line (2021 on): books 0.657 (61.6% right), our model 0.667 (58.1%). Where a total is posted, the pages show the books' over/under chance.
For the record: flat bets wherever the blend was 3+ points above the closing line would have returned +1.8% per bet over 2,094 bets since 2016, priced at the middle of the closing range across books. That is within what luck produces over this many bets and assumes prices most bettors couldn't get, so it is not evidence of an edge, and we don't claim one.
Are the percentages honest?
| We said | Fights | Average call | Actually won | |
|---|---|---|---|---|
| 50%-60% | 1625 | 55% | 57% | |
| 60%-70% | 1713 | 65% | 63% | |
| 70%-80% | 1286 | 75% | 75% | |
| 80%-90% | 650 | 84% | 87% | |
| 90%-100% | 74 | 92% | 91% |
Year by year (log loss, lower is better)
| Year | Fights | Market | Blend | Ours alone |
|---|---|---|---|---|
| 2016 | 483 | 0.6246 | 0.6204 | 0.6532 |
| 2017 | 444 | 0.6119 | 0.6059 | 0.6507 |
| 2018 | 468 | 0.6007 | 0.5965 | 0.6464 |
| 2019 | 503 | 0.6287 | 0.6283 | 0.6639 |
| 2020 | 443 | 0.6157 | 0.6146 | 0.6550 |
| 2021 | 497 | 0.6206 | 0.6164 | 0.6475 |
| 2022 | 506 | 0.6024 | 0.5987 | 0.6464 |
| 2023 | 504 | 0.6014 | 0.6014 | 0.6497 |
| 2024 | 513 | 0.5813 | 0.5767 | 0.6314 |
| 2025 | 512 | 0.5887 | 0.5839 | 0.6257 |
| 2026 | 475 | 0.5848 | 0.5884 | 0.6516 |
Data: ufcstats.com via Greco1899/scrape_ufc_stats; closing lines from the UFC Data dataset (Bair, N., 2026, CC BY 4.0); cards, results and DraftKings lines from ESPN's public fight center.