Left side is the play. Right side is how often we think it hits. That number is our model’s probability pulled partway toward the book’s own implied number, so it is not raw model output.
The card is ranked by edge, not by percentage: how far our number sits above what the price implies. A 68% play at -240 has no edge, because -240 already implies about 70%. Nothing priced past -175 makes the card.
Every NHL play clears the same bar and shows the same way. There is no strong tier, because higher numbers did not hit more often when we tested the model on last season.
Max 5 plays a day and 3 per board, so a play that clears the bar can still miss on a busy day.
The parlay builder shows each leg’s probability across every market, including alt lines, totals and home runs we don’t back straight. So a builder leg can show a higher number than anything here and still not be a Top Play. The builder shows how likely a leg is. Top Plays shows where the price is wrong.
Postseason numbers run lower on purpose: starters go shorter and lineups are the best nine every inning.
Who is out tonight moves these numbers more than anything else. The chips on a card say when a big-minutes teammate is questionable, when tonight's projected starting five differs from the last game's, and when a heavy-minutes player is on the second night of a back-to-back.
NOT LIVE. This card is in testing and nobody else can reach it. Numbers are not final until the models clear their replay gate.
SYNTHETIC LINES. No book posted these lines. They are hung by the sharp-book baseline so the whole loop can run; the record from them is not meaningful.