Verdicts, from Sept 28.MLB:STRONG BET = model edge 5+ over Robinhood's price and either the crowd's money is on the other team or it is one of the day's most-bet games (12-2 there). LEAN = edge 2+ otherwise; about break-even, small stakes or skip. FADE (dog) = the favorite is priced about right (edge -3 to +2), so take the underdog at the price shown. BET (chalk) = the model thinks the favorite is overpriced by 3+, but those favorites keep winning: bet the favorite. NFL (rebuilt Oct 16, weeks 2-6): the signal is the spread market, not the moneyline crowd. The team with less of the crowd's spread money won 30 of 42 games outright; the team with more won 12 of 42. STRONG · dog = 🌙 primetime (Thursday, Sunday night, Monday) underdog when the crowd is on the favorite: 10-0 in weeks 3-6. STRONG BET = Sunday-daytime favorite with less of the spread money, in a spread market of $300k+: 8-0. BET = the same in a $100-300k market: 4-2. Sunday underdogs are a PASS on the moneyline even with less spread money (7-10): take their spread on the Lines page instead. PASS (wait) = under $100k of spread money so far; the side has flipped before kickoff at that size, so check back closer to kickoff. Samples are small (42 games with spread data). Every older rule stays in the scorecard below so they keep being compared. Pick = the side the model likes against Robinhood's price. Model = win chance for that side. Robinhood = what a $1 contract on that side costs right now (the ask), with the equivalent American odds; Robinhood's contracts are Kalshi's. Edge = model minus that price, in points; your fee is about a penny a contract, so +2 is real.
Public $ on pick = share of the Kalshi/Robinhood dollars on the pick side: over 65% is a crowded side. $ at risk = open interest at lock: contracts still held, so the dollars actually riding on the result, with no churn (hover for the raw traded volume, which counts both sides of every trade and re-trades; a ~ means only volume was recorded for that game). Tape $ on pick = directional money from Kalshi's pre-game trade tape: every trade credited to the team the aggressor bet on (YES on a team, or NO on its opponent); hover for the dollars. Tape $ = total taker dollars before start. Unlike Public $ on pick, which counts both sides of each market's volume, this one says which team the money actually backed.
Book take = once a game is final, the dollars the crowd put on the losing team, from the trade tape (the same source as Public $ on pick); the tile sums it for the slate. Robinhood/Kalshi is an exchange, so this is what the winning bettors collected from the losing ones, not a house profit. Units = flat 1u on every model pick at Robinhood's price. Book gave = what the winning bettors were paid: their profit at the price they bought (the dollars on the winning team x (1 - price) / price; $102k on a 61¢ / −156 favorite wins about $65k, not $102k), using the closing ask. Pinnacle = the sharpest book's de-vigged chance (a ~ means Pinnacle hasn't posted yet, so it's the Bovada/BetOnline average until it does).
There are three different "favorites" on every game and they do not agree: the side the model has over 50%, the side Robinhood prices over 50¢, and the side the public's money is on. Only the first one predicts anything. Weekend 1, 28 graded games at Robinhood prices: model's side over 50% went 9-5 (+1.3u); the market's priced favorite went 7-5 but lost 1.0u (short prices); the public's side went 8-8 and lost 2.0u.
The Pick is always the model's favorite, its side over 50%. FADE (dog) = the pick is a favorite priced about right (edge -3 to +2, a PASS), so take the other side at the price shown: those dogs are 25-21 for +15.5u in MLB and 10-15 for +7.6u in NFL. When the favorite looks overpriced by 3+ the model is usually the one that is wrong (dogs 9-29 in MLB), so those are a plain PASS. 🔥 most-bet marks the slate's most-bet games by $ at risk. It is a flag, not a verdict: dogs in those games won 16 of 33 overall, but almost all of those were dogs the model itself picked; when the model has the favorite the dog is 10-23. The scorecard tracks it. MLB verdicts key off the Edge column, the model's win % minus Robinhood's price: STRONG BET = edge of 5 or more, BET = edge of 2 to 5, PASS = under 2. All of the MLB profit so far has come from the 5+ games; the 2 to 5 games have only broken even. NFL verdicts: see the rebuilt NFL rules above (spread money, not the moneyline edge). For NFL, Public $ on pick comes from the trade tape (each pre-game trade credited to the team the bettor actually backed), because splitting volume between the two team contracts badly understates the money on big favorites; a ~ means no tape yet, so the volume split is shown. MLB still uses the volume split. The NFL rule is still unproven. Ignore what Robinhood or the crowd calls the favorite; bet the model's side, preferably when the crowd isn't there.
MLB model: regressed run-differential strength, starting-pitcher runs-allowed adjustment, home field. NFL model: last season's point differential (regressed) plus 2 points for home; weak until 2026 games exist, so lean on Pinnacle vs Kalshi there.