PROPULATION NFL
— loadingWEEK 1
PACK
Three overs and three unders, ranked on edge × measured calibration × book count — a 6-point edge on a well-calibrated prop outranks a 10-point edge on a shaky one.
Disagreements above 25 points are held back: outside the range the model's calibration was measured in, they have always turned out to be an input disagreement rather than value.
PAPER MODE — beats a naive benchmark, not yet the market. What that means ›
Measured to beat a naive benchmark (each player's own trailing mean) on 2021–24. It has not been shown to beat the market: gate G5 needs a settled line history, which only began accumulating 2026-09-10. Treat every edge as paper trading and watch closing-line value rather than the edge itself.
Compare Med against the line, not Proj. Yardage outcomes are right-skewed (measured mean/median 1.06–1.16), so a 50/50 line sits below the mean by design. The edge column already uses the full distribution, not either single number.
| Player | Pos | Matchup | Prop | Proj | 80% band | Med | Line | Model | Market | Edge | Side | Conf | Bks |
|---|
WEEK 1
PACK
The six plays where the model disagrees most with a PrizePicks line, ranked on P(hit) × measured calibration.
A PrizePicks pick is a straight 50/50 proposition, so the edge is simply how far the model's probability sits from 50% — there is no vig to strip. Standard lines only: demons and goblins are shaded on purpose and are not 50/50.
How to read this board — and what it does not claim
The price edge is a fact and needs no model. Kalshi is an exchange: it charges a trading fee instead of building a margin into the quote, so its ask for a given outcome usually sits below the sportsbook price for the same outcome. Measured across 4,353 rungs that both Kalshi and at least one US book quoted on this slate, Kalshi was cheaper on 66% of them. On liquid, parlay-usable rungs the average advantage was 1.2 cents against the single best book price and 3.1 cents against the typical one — and that advantage compounds multiplicatively across a combo.
Receptions is the best prop for this, and it is not close. 78% of Kalshi's reception rungs beat the best book, and 68 of the 120 largest edges on the board are reception rungs. Rushing attempts and interceptions are excluded outright: Kalshi is measurably the worse venue on those two (−0.15 and −0.23 cents).
"EV vs market" is negative on purpose. Buying at the ask means crossing the spread, and the exchange's own mid is the only vig-free reference available. Every row here is slightly negative against it — the same bet at a sportsbook is considerably more negative. This board finds the cheapest execution of a parlay; it does not claim the parlay is free money.
"Model edge" is the unvalidated column. The model's own probability comes from a simulation that draws a whole game at once, so a same-game combo's true joint probability is read off the draws rather than assumed independent — which is genuinely more than a sportsbook screen can tell you. But gate G5, the market-beat regression, has never run. Treat the model column as an opinion and the price column as arithmetic. Any leg where the model disagrees with the market by more than 20 points is dropped outright: those are depth-chart disputes, and the books win those far more often than a projection does.
The combo markup is measured, and it is not a constant. Combos are priced by request-for-quote (their public order books never fill), so we asked makers to quote six real combos and compared each answer with the exact product of its legs' asks fetched at the same moment. Three quoted on Saturday afternoon came back +2.3%, −2.1%, +2.1% over the leg product; three quoted late Saturday night, hours before kickoff, came back +6.7%, +6.9%, +8.6%. Makers price a combo near the product of the leg mids plus their own spread, and how wide that spread is depends on the hour. Cards are priced at the six-point mean (+4%) and the range beside each payout spans the full observed spread (−2% to +9%) — it is wide because the thing it measures moves. The "get real quote" button asks the makers right now, and that number is the price. Trading fees are estimated from the published per-contract formula and remain unverified for combos.
"$100 pays" at the leg product is exact. That figure comes from Kalshi's own four-decimal asks, not from rounded display percentages, and it is the number to trust — the markup is a cost stacked on top of it. It also does not touch the edge claim: at the worst observed markup of 10%, a ~15¢ saving on a ~47¢ ticket still clears ~10¢.
The "save" column compares leg prices, not combos. Leg product to leg product, with neither venue's parlay machinery involved — that isolates the thing actually being measured. A sportsbook parlay of the same legs carries its own additional hold on top of its leg prices, larger than Kalshi's and not measured here, so the comparison is if anything generous to the book.
Liquidity is real and it binds. "Max stake" is the smallest depth-at-the-ask across the legs. Past that number you are moving the price and the edge shrinks.