The goal is to make informed picks, show the work, and learn from the results. Confidence needs to be supported by the record.
Confidence estimates how likely a pick is to cover, excluding pushes. Expected cover margin estimates how far the final score will finish above or below our spread, on average. A team can be more likely to cover without being expected to win by a larger margin.
Under this proposed scale, 5.6 would mean a 56% chance of covering, excluding pushes. The earlier 6.0–8.5 ratings were judgments, not win probabilities. They stay separate.
The starting point is the spread and price across sportsbooks at the time of the pick. The proposed model would estimate the possible final scoring margins, then assess player availability, matchups, rest, travel, and line movement. An adjustment needs evidence that it helps on games outside the data used to build it.
Football scores cluster around key margins, so the calculation needs to account for those outcomes and pushes. A difference between ticket and money percentages can prompt a closer look. It does not establish who has better information.
College and NFL need separate testing. The proposed process uses earlier games to build the model and later games to test it. Record the forecast, spread, price, and model version before kickoff. Account for picks that share the same week or matchup when measuring uncertainty.
Calibrate probabilities on a separate period, then compare them with market probabilities after removing the sportsbook margin. This is out-of-sample testing. The original picks and ratings stay as published.
The review should cover wins and losses, the margin against our published spread, and the verified Wynn close. For future probability forecasts, use Brier score and calibration to check whether the stated confidence matches results. For projected margins, measure average absolute error, directional bias, and how often outcomes fall within the forecast range.
Show how many picks support each calculation and what information is missing. Win rate alone does not establish a return. The price matters. Pushes return the stake, and missing prices remain unknown.
56% cover probability; 44% loss probability.
+0.069
Break-even: 52.4%. Move the slider to see how the calculation works. This example is not a forecast for any published pick.
This compares the ratings published before kickoff with the results against our spread. It helps us review the earlier judgments without changing them after the game.
Above zero beat the posted spread; below zero missed it. Average across all wins and losses, not winners alone.
Original editorial ratings · 15 settled picks · no ratings backfilled. Inspect the sample
For new, validated forecasts: score = 10 × P(cover | no push). A 5.6 represents a 56% estimated cover probability, not a six-point cover. Historical editorial ratings are kept separate.
Expected cover margin would be a separate forecast. Both need testing on future games, with the original timestamp and the full record kept intact.