Methodology

Prices

All prices are live prediction-market quotes read at publish time, shown in cents (a 45¢ price = 45% implied). If a market has no quote the game is marked no price; prices are never estimated, except parlay prices, which are the product of the leg asks and labeled as estimates.

NHL picks

Win probability = logistic model on: score-adjusted expected-goals share (MoneyPuck, all situations), starting-goalie goals saved above expected per 60 (shrunk toward zero for small samples), a home-ice term, and a back-to-back penalty from the prior night's NHL schedule. Starters come from daily starting-goalie reports; picks with unconfirmed starters are capped at 3/5. The puck line (−1.5) cover rate is ~57% of the favorite's win probability. Each game gets the side with the largest edge over the price (ties favor the moneyline). Confidence 1–5 scales with edge.

NFL / CFB plays

Model probabilities versus market no-vig prices and sportsbook consensus, with injury, weather and line-move checks. Only rated plays (4/5+) are listed.

College football picks

Team strength = average of ESPN's Football Power Index and our own margin rating fitted to every 2026 FBS result (margins capped at 28, home field fitted). The fair margin blends 35% model with 65% sportsbook consensus spread; win and cover chances assume a normal margin spread of 15.5 points. Each game gets the moneyline or main-line spread with the higher expected value. Confidence scales with edge, drops when model and market differ by more than 10 points or no sportsbook line exists, and is capped at 3/5 until graded.

Anytime TD

Expected team touchdowns from market-implied points, split by each player's red-zone and end-zone opportunity share, adjusted for the opponent's TDs allowed to that position. Not yet backtested.

Prop parlays

Alt lines the player has cleared in every (or nearly every) game this season, filtered for injuries and matchup. Legs are treated as independent.

Grading & CLV

Picks are logged with timestamp, price, model % and confidence when first published, and graded from official final scores and box scores. ROI is at 1 unit flat. CLV compares the publish price with the last pre-game price.

Caveats

The models are early and small-sample. In backtests the market has been better calibrated than the model; treat everything here as research.