Research center · Compare models · Full NFL results · Methodology
NFL performance evidence
Completed official Play/Strong selections with matching original pre-kickoff receipts. Historical rows without matching receipts remain in the full history but are excluded here.
Small verified sample: no predictive superiority or profitability is established.
Full official history: 22 completed · Excluded from this receipt-backed sample: 11 · Without an original receipt: 11. Inspect every completed stored pick and its receipt status.
One unit risked per selection at assumed -110; pushes return the stake. Not actual-price returns.
Results by market
| Group | Completed | Record | Win rate | Assumed −110 ROI | CLV coverage |
|---|---|---|---|---|---|
| SPREAD | 5 | 4-1 | 80% | 52.7% | 1/5 |
| TOTAL | 6 | 2-4 | 33.3% | -36.4% | 3/6 |
Results by model version
| Group | Completed | Record | Win rate | Assumed −110 ROI | CLV coverage |
|---|---|---|---|---|---|
| 4.5.1 | 9 | 4-5 | 44.4% | -15.2% | 3/9 |
| 5.0.0 | 2 | 2-0 | 100% | 90.9% | 1/2 |
Weekly verified record
Only completed picks with matching pregame receipts. A week with no eligible receipts is not shown as 0–0. Spread and total results stay separate.
| Season / week | All picks W–L–P | Win rate | Spread W–L–P | Total W–L–P | Model versions |
|---|---|---|---|---|---|
| 2026:W4 | 3-4 | 42.9% | 2-1 | 1-3 | 4.5.1 |
| 2026:W5 | 2-0 | 100% | 2-0 | 0-0 | 5.0.0 |
| 2026:W6 | 1-1 | 50% | 0-0 | 1-1 | 4.5.1 |
Closing-line value: show the coverage
4 of 11 verified picks have usable recorded closing lines; 7 are missing. Missing values never count as zero.
Spread CLV: -0.5 points across 1 picks. Total CLV: 0 points across 3 picks. Positive values mean a better saved line. These are line-point comparisons, not price-adjusted expected returns. Legacy zero closing lines remain excluded as ambiguous placeholders.
Probability accuracy
0 decided picks have a probability preserved on the original receipt; 11 otherwise verified decisions lack that evidence. No probabilities are reconstructed after the game.
Brier score: — · Log loss: — · Constant 50% Brier baseline: —. Lower scores are better.
Only probabilities preserved on the original pregame receipt. Pushes excluded; log loss clips probabilities to [1e-15, 1-1e-15]. The 50% baseline is descriptive, not a sportsbook forecast.
| Forecast band | Decisions | Average forecast | Actual win rate |
|---|---|---|---|
| 0–50% | 0 | — | — |
| 50–55% | 0 | — | — |
| 55–60% | 0 | — | — |
| 60–65% | 0 | — | — |
| 65–70% | 0 | — | — |
| 70–100% | 0 | — | — |
What would demonstrate an advantage?
A growing forward sample, reliable inputs, probability calibration and comparisons with the same pregame sportsbook lines. This report does not establish superiority over sportsbooks or other analytics sites. Actual-price ROI and a paired market-probability benchmark require saved executable prices and are not inferred from point spreads or assumed −110 returns.
Evidence summary (JSON) · Full completed history (CSV) · Historical performance study
Updated 2026-10-18T19:07:18.038Z. Official picks, experimental comparisons and non-official leans remain separate.