72.5%
Directional Accuracy
255 games · 2021–2024
+30.5%
ROI on Upset Picks
62 bets · +18.9u net
56.5%
Bet Accuracy
35 of 62 upset picks correct
+0.46
Avg Closing Line Value
Consistent edge vs market
// Year-by-Year Results
| Year | Games | Accuracy | Bets | Bet Accuracy | ROI | Condition |
|---|---|---|---|---|---|---|
| 2021 | 66 | 68.2% | 14 | 42.9% | −1.5% | ⚠️ Tight gaps · avg 10.1 |
| 2022 | 63 | 71.4% | 17 | 58.8% | +35.7% | ✅ Moderate gaps |
| 2023 | 63 | 73.0% | 16 | 56.2% | +30.0% | ✅ Moderate gaps |
| 2024 | 63 | 77.8% | 15 | 66.7% | +55.2% | 🔥 Wide gaps · avg 12.0 |
| Total | 255 | 72.5% | 62 | 56.5% | +30.5% | 4-year mean |
// Results by Round
| Round | Accuracy | Bets | ROI |
|---|---|---|---|
| Round of 64 | 72.4% | 32 | +6.2% |
| Round of 32 🔥 | 78.5% | 14 | +42.9% |
| Sweet 16 ⚠️ | 50.0% | 9 | +11.1% |
| Elite 8 | 75.0% | 5 | +20.0% |
| Final Four | 75.0% | 1 | — |
🔥 R32 is the highest-signal round. S16 accuracy drops as elite teams dominate.
// Results by Seed Matchup
| Matchup | Games | Accuracy | ROI |
|---|---|---|---|
| 4v5 🔥 | 9 | 88.9% | +50.0% |
| 6v11 🔥 | 16 | 62.5% | +42.9% |
| 8v9 🔥 | 16 | 75.0% | +38.5% |
| 3v6 | 7 | 71.4% | +20.0% |
| 7v10 ❌ | 15 | 40.0% | −33.3% |
| 1v16 | 16 | 93.8% | No bet |
7v10 is the only consistently unprofitable matchup. 1-seeds: no bet, no edge needed.
// About the Model
Our E12-3 Gap-Aware model uses Barttorvik adjusted efficiency margins as the primary signal with gap threshold calibration and defensive adjustments. Tested on 255 games across 2021–2024, validated against multi-year ensemble testing. The model is selective — it only flags ~24% of games as high-confidence picks, which is where the edge lives. Accuracy varies with tournament gap distribution: wide-gap years (2024) produce higher accuracy than tight-gap years (2021). 2026 projection: 70–78% directional accuracy depending on bracket characteristics.