Accuracy scoreboard
Proof for the model itself. Every finished game is re-projected using only what was known before that week, then compared with what actually happened (PPR).
Live track record
Every morning, before kickoff, the projections are saved and locked. After the games, the saved numbers are compared with what happened. Nobody can edit them afterward, so this is the honest scoreboard.
The matchup model beat a plain season average by 0.4 points per player-game. Players the model liked most scored 8.4 more points than the ones it liked least. Correlation between projection and result: 0.46. Average bias: +0.3 points (positive = projections ran high).
Early in the season this is a small sample and Vegas, weather and injury adjustments are not part of the backtest, since those need historical snapshots. It will get more convincing every week.
| Position | Tested | Model miss | Avg miss | Corr |
|---|---|---|---|---|
| QB | 201 | 6.6 | 7.0 | 0.26 |
| RB | 303 | 6.2 | 6.6 | 0.44 |
| WR | 492 | 5.9 | 6.3 | 0.44 |
| TE | 220 | 5.4 | 6.0 | 0.37 |
| K | 194 | 4.0 | 4.0 | 0.11 |
Bet markets
Are the Bets page percentages honest? Every common line for every player who played is re-priced using only what was known before that week, then checked against what happened. A good model's 60% calls hit about 60% of the time.
When we said… it happened
| We said | Calls | Avg chance | Hit |
|---|---|---|---|
| 0–10% | 7044 | 3% | 6% |
| 10–20% | 2427 | 15% | 19% |
| 20–30% | 1840 | 25% | 28% |
| 30–40% | 1438 | 35% | 34% |
| 40–50% | 1094 | 45% | 43% |
| 50–60% | 865 | 55% | 51% |
| 60–70% | 742 | 65% | 58% |
| 70–80% | 597 | 75% | 66% |
| 80–90% | 476 | 85% | 74% |
| 90–100% | 303 | 94% | 85% |
By bet type
| Market | Calls | Avg chance | Hit | Error vs naive |
|---|---|---|---|---|
| Anytime touchdown | 2706 | 12% | 12% | 0.087 / 0.093 |
| Rushing touchdown | 1152 | 12% | 13% | 0.089 / 0.094 |
| Receiving touchdown | 2220 | 10% | 9% | 0.069 / 0.076 |
| Passing touchdowns | 504 | 47% | 47% | 0.184 / 0.189 |
| Receptions | 3700 | 36% | 34% | 0.180 / 0.169 |
| Rushing yards | 1185 | 31% | 34% | 0.176 / 0.181 |
| Field goals | 519 | 52% | 56% | 0.217 / 0.201 |
| Receiving yards | 3700 | 23% | 26% | 0.161 / 0.156 |
| Passing yards | 840 | 37% | 41% | 0.213 / 0.211 |
| Quarterback rushing yards | 300 | 25% | 23% | 0.170 / 0.166 |
Greyed markets have fewer than 100 graded lines, too few to judge. Injury and quarterback-change adjustments aren't in this test (they rely on reports we only have for today), so it grades the core model.
SP Score
Does a higher SportsProof Score mean more yards, catches and touchdowns the next game? Each finished week from Week 4 on is re-scored using only earlier games. “Rank match” is how well the score orders players by what they then did (within position, 0 = no better than random); it is compared with simply using each player's season average.
| Bet type | Scored | Rank match | Season avg | Top vs bottom fifth |
|---|---|---|---|---|
| Anytime touchdown | 762 | 0.23 | 0.16 | 46% vs 16% |
| Rushing touchdown | 319 | 0.34 | 0.18 | 63% vs 11% |
| Receiving touchdown | 639 | 0.11 | 0.07 | 30% vs 13% |
| Passing touchdowns | 144 | 0.26 | 0.22 | 59% vs 29% |
| Receptions | 639 | 0.45 | 0.44 | 65% vs 10% |
| Rushing yards | 196 | 0.48 | 0.54 | 55% vs 13% |
| Field goals | 141 | -0.01 | 0.06 | 66% vs 54% |
| Receiving yards | 639 | 0.38 | 0.35 | 51% vs 14% |
| Passing yards | 144 | 0.29 | 0.28 | 62% vs 18% |
| Quarterback rushing yards | 82 | 0.40 | 0.23 | 47% vs 6% |