1678-951-10 · n=2629
Straight-up winners · 2016–2025
Who won the game?
The 2016–2025 regular seasons are retrospective development results, not a clean holdout, against a 70% target. The model called the winner in 1678-951-10 games (63.8%, n=2629). Betting the favorite on those games won 66.7% (n=2629). The model finished below that baseline by 2.9%. High-confidence calls, model win probability at least 65%, were 72.5% (n=1027). The favorite won 72.9% of those same games, and the model agreed with the favorite 97.5% of the time. The point estimate is above 70% because those games are lopsided favorites, not because the model beat the favorite. No individual season from 2016 through 2025 reached 70% for the model. The 10-season overall rate does not clear 70%.
Same games with a favorite · n=2629
745-282-5 · model P ≥ 65% · n=1027
A winner call is the side with the higher model win probability. Ties are excluded from the hit rate. The favorite is the side with the higher moneyline implied probability, or the spread favorite when the moneyline is missing. Pick'ems are not a favorite. Tiers are model win probability: lean 50–58%, medium 58–65%, high 65% and up. Those cuts were set before the 2016–2025 results were read.
By season
2026 is the season to date and is not inside the 10-season total.
| Season | Record | Model | Favorite | Gap | n |
|---|---|---|---|---|---|
| 2016 | 154-100-2 | 60.6% | 63.8% | -3.1% | 254 |
| 2017 | 171-85-0 | 66.8% | 71.5% | -4.7% | 256 |
| 2018 | 161-93-2 | 63.4% | 66.5% | -3.1% | 254 |
| 2019 | 163-92-1 | 63.9% | 64.3% | -0.4% | 255 |
| 2020 | 173-82-1 | 67.8% | 67.8% | 0.0% | 255 |
| 2021 | 169-102-1 | 62.4% | 62.4% | 0.0% | 271 |
| 2022 | 171-98-2 | 63.6% | 65.8% | -2.2% | 269 |
| 2023 | 165-107-0 | 60.7% | 68.0% | -7.3% | 272 |
| 2024 | 183-89-0 | 67.3% | 71.7% | -4.4% | 272 |
| 2025 | 168-103-1 | 62.0% | 65.3% | -3.3% | 271 |
| 2026 | 94-55-1 | 63.1% | 67.8% | -4.7% | 149 |
By confidence
Lean is a 50–58% model win probability, medium 58–65%, high is 65% and up. The favorite column is the favorite’s win rate on the same games.
480-419-3 · favorite 60.2% · n=899
453-250-2 · favorite 66.0% · n=703
745-282-5 · favorite 72.9% · n=1027
Raising the probability cutoff shrinks the number of calls.
| Min probability | Model | Favorite | n |
|---|---|---|---|
| 50% | 63.8% | 66.7% | 2629 |
| 55% | 67.2% | 68.7% | 2059 |
| 60% | 71.0% | 71.9% | 1518 |
| 65% | 72.5% | 72.9% | 1027 |
| 70% | 76.4% | 77.0% | 640 |
On 2639 out-of-sample games, the actual margin regressed on the closing spread and the model-minus-line residual gives the residual a coefficient of 0.011 (SE 0.088, p 0.905). A coefficient near zero means the model is not adding a detectable margin signal beyond the close.
Bankroll simulator
Regular season only, 2016 through 2026 to date, after recent form was removed from the margin. Winner calls are the side with the higher model win probability, settled at that side's closing moneyline. High confidence is a model win probability of at least 65%. Spread and total use the published 3% edge picks. Spread prices use the wider 2015 sigma, and weeks 1–4 are not treated as high confidence. Flat $10 is also 2% of the $500 start. Compounding stakes 2% of the balance after the previous bet, rounded to the cent. A flat bet is skipped, and the path stops, when the balance cannot cover $10.
| Path | Ending | P/L | ROI | Max drawdown | Low | Bets |
|---|---|---|---|---|---|---|
Moneyline winner calls, flat $10 Stopped 2021-11-14: balance $5.67 could not cover the next $10 | $5.67 | -$494.33 | -3.5% | $494.3398.9% | $5.67 | 1425of 2785 |
High-confidence winner calls, flat $10 Played every offered bet | $100.15 | -$399.85 | -3.7% | $500.3194.7% | $27.96 | 1092 |
Spread picks, flat $10 Stopped 2023-12-03: balance $3.97 could not cover the next $10 | $3.97 | -$496.03 | -3.6% | $642.4999.4% | $3.97 | 1358of 1862 |
Total picks, flat $10 Stopped 2019-11-24: balance $3.92 could not cover the next $10 | $3.92 | -$496.08 | -6.8% | $592.7399.3% | $3.92 | 733of 2271 |
Always the moneyline favorite, flat $10 Stopped 2021-11-14: balance $6.30 could not cover the next $10 | $6.30 | -$493.70 | -3.5% | $587.3398.9% | $6.30 | 1418of 2772 |
Moneyline winner calls, 2% of bankroll Played every offered bet | $39.88 | -$460.12 | -4.4% | $463.0492.6% | $36.96 | 2785 |
High-confidence winner calls, 2% of bankroll Played every offered bet | $207.05 | -$292.95 | -3.7% | $349.1966.0% | $179.75 | 1092 |
Spread picks, 2% of bankroll Played every offered bet | $67.91 | -$432.09 | -3.8% | $614.6392.2% | $52.31 | 1862 |
Total picks, 2% of bankroll Played every offered bet | $21.18 | -$478.82 | -6.5% | $575.2696.6% | $20.13 | 2271 |
Main path: Moneyline winner calls, flat $10
Balance at the end of each season. Flat paths stay put after they can no longer bet $10.
Highest ending balance: High-confidence winner calls, 2% of bankroll
Balance at the end of each season. Flat paths stay put after they can no longer bet $10.
Secondary: against the number
These 2016–2025 rows are retrospective development results. The forward record is on the paper ledger. Spread picks leave out recent form, use the 2015 sigma scale, and label weeks 1–4 as caution. Break-even at −110 is 52.4%. The pre-registered bar is a 70% hit rate. Break-even at standard -110 odds is 52.4%. No regular-season slice with at least 200 decided bets reached 70%. No thinner slice printed a 70% point estimate either. Spread at the 3% edge threshold is 889-925-48 (49.0% hit, -4.7% ROI, n=1814). Moneyline at the 3% edge threshold is 754-1271-10 (37.2% hit, -8.1% ROI, n=2025). Total at the 3% edge threshold is 1095-1159-13 (48.6% hit, -5.8% ROI, n=2254). High-confidence spread (edge at least 8%) is 274-281-9 (49.4%, n=555). High-confidence moneyline (edge at least 8%) is 340-649-4 (34.4%, n=989). High-confidence total (edge at least 8%) is 681-745-8 (47.8%, n=1426). Spread volume falls from 2232 bets at any positive edge to 742 bets at an 8% edge. Player-prop ROI against closing lines is not in this report: the free data has no historical prop prices.
spread
Bars are bet count. Dots are hit rate. Green tick is 52.4%. Gold tick is 70%.
moneyline
Bars are bet count. Dots are hit rate. Green tick is 52.4%. Gold tick is 70%.
total
Bars are bet count. Dots are hit rate. Green tick is 52.4%. Gold tick is 70%.
Confidence tier
Lean is 3–5 points of edge, medium 5–8, high is 8 or more. High-only is the right-hand column.
| Market | Tier | Record | Hit rate | ROI | n | 95% interval |
|---|---|---|---|---|---|---|
| spread | lean | 233-254-15 | 47.8% | -7.0% | 487 | 43.4%–52.3% |
| spread | medium | 285-300-20 | 48.7% | -5.0% | 585 | 44.7%–52.8% |
| spread | high | 274-281-9 | 49.4% | -4.0% | 555 | 45.2%–53.5% |
| moneyline | lean | 167-265-5 | 38.7% | -11.0% | 432 | 34.2%–43.3% |
| moneyline | medium | 247-357-1 | 40.9% | -6.2% | 604 | 37.0%–44.9% |
| moneyline | high | 340-649-4 | 34.4% | -8.0% | 989 | 31.5%–37.4% |
| total | lean | 161-188-1 | 46.1% | -10.5% | 349 | 41.0%–51.4% |
| total | medium | 253-226-4 | 52.8% | 2.3% | 479 | 48.3%–57.3% |
| total | high | 681-745-8 | 47.8% | -7.3% | 1426 | 45.2%–50.3% |
By season
Regular season, 3% edge. 2026 is the season to date, not a full year.
| Season | spread | moneyline | total |
|---|---|---|---|
| 2026 | 53.8%49-42-5 · 3.5% ROI | 44.2%46-58-1 · 8.0% ROI | 50.8%62-60-0 · -3.0% ROI |
| 2025 | 42.7%76-102-1 · -17.8% ROI | 38.4%78-125-1 · -6.3% ROI | 51.9%123-114-0 · -0.6% ROI |
| 2024 | 49.4%86-88-2 · -4.9% ROI | 38.0%73-119-0 · -15.6% ROI | 52.1%113-104-3 · -0.5% ROI |
| 2023 | 46.2%79-92-10 · -11.1% ROI | 33.7%65-128-0 · -14.4% ROI | 44.2%92-116-2 · -15.5% ROI |
| 2022 | 48.6%85-90-7 · -5.0% ROI | 35.2%68-125-2 · -15.0% ROI | 47.0%103-116-2 · -8.5% ROI |
| 2021 | 45.5%85-102-3 · -11.4% ROI | 33.7%65-128-1 · -9.8% ROI | 43.3%107-140-2 · -16.0% ROI |
| 2020 | 52.5%93-84-0 · 2.4% ROI | 36.5%69-120-0 · -16.2% ROI | 53.2%108-95-3 · 4.0% ROI |
| 2019 | 52.3%80-73-8 · 2.1% ROI | 37.7%69-114-1 · -6.8% ROI | 50.0%105-105-0 · -2.3% ROI |
| 2018 | 51.7%92-86-5 · 0.7% ROI | 38.0%73-119-2 · 3.3% ROI | 48.7%95-100-1 · -4.5% ROI |
| 2017 | 50.9%81-78-4 · -0.0% ROI | 35.9%69-123-0 · -6.3% ROI | 42.4%87-118-0 · -16.6% ROI |
| 2016 | 48.5%83-88-3 · -4.1% ROI | 41.4%79-112-2 · -2.8% ROI | 52.4%100-91-0 · 2.8% ROI |
Calibration
Each dot is a bin of predicted probability against how often that side actually won. A calibrated model sits on the diagonal. ECE is the average gap.
spread
ECE 0.063 · n=2718
moneyline
ECE 0.026 · n=2777
total
ECE 0.106 · n=2768
Always betting the home spread covered 48.6% of 2718 decided lines.
A fair closing number lands near 50%. Margin MAE 10.15 · total MAE 11.11.
spread: 42-38-0 · 52.5% · n=80
moneyline: 33-57-0 · 36.7% · n=90
total: 47-39-2 · 54.6% · n=86