Weekly and rest-of-season projections from the league's own model — whole distributions, not point guesses. Floors, ceilings, and how likely a player is to blow up.
| Consulting the crystal ball… |
Every number here is a draw from a probability distribution the model assigns to each player-game, not a single guess. Mean is the expected score; Median the 50/50 line; p10 and p90 are a realistic floor and ceiling (the outcome lands between them 80 % of the time); P>20 is the chance of a 20-plus-point week. Weekly numbers include the season layer: Play % is the chance the player suits up (from this week's injury designation, then the season's injury chain for later weeks — pick any remaining week from the Week menu), Mean counts a missed game as zero, and If plays is the expectation if he does. Switch the columns to Stat line for the expected box score behind the points (attempts, yards, touchdowns, targets, receptions), and sort any column by clicking its header — Mean is points per game in the weekly view, and PPG is points per game played for the rest of the season. On the chart, further right is a better projection and higher up is a wider range of outcomes — a player high and right is a boom-or-bust bet, one low and right is a safe start. "Relative" volatility divides the spread by the projection so cheap and expensive players compare fairly.
Rest-of-season totals come from 10,000 simulated seasons that include the chance of missed games (Exp. games is the average number actually played) and the week-by-week dependence of a player's form. Top-12 is the share of simulated seasons in which the player finishes inside the top twelve at his position from here on.
The Props view puts live sportsbook player props (passing, rushing and receiving yards, receptions, passing touchdowns, anytime touchdown) next to the model. For every line the book's price is turned into a vig-free probability, the model's chance of clearing the same line is read off that stat's distribution, and Edge is the difference (positive means the model likes the over). The chart plots book against model: points far from the diagonal are where the two disagree. Click any player in the weekly table to open the stat-by-stat distribution with its props underneath.
FanDuel and DraftKings props are recorded on game days at 9:00 ET, for the games kicking off that day (to save API credits). In the Props view, Open is the line when a prop first appeared and Move, Book Δ and Model Δ show how the line, the book's vig-free probability and the model's probability have shifted since; with one pull per game these stay flat unless a game was pulled more than once. The Record view grades every prop once its game is final: how often the model's side won, what flat bets on the model's edges would have returned at the opening and closing prices, how often the market later moved toward the model (beating the close is the best sign an edge is real), and whether the model's probabilities are as well calibrated as the books' (lower Brier score is better). Props where the player did not play are void, as at the books.
The Lineups view builds every fantasy team's best possible starting lineup for the week from the projected means (QB, two RB, three WR, TE, two FLEX, one Superflex; the defense slot is not modeled), using the rosters Sleeper reports right now. The bars rank the league by projected optimal total; pick a team with the Owner filter (or click its bar) to see its lineup slot by slot, with the bench below and close calls — bench players within a point and a half of a starter — highlighted in gold. In the Rest of Season view, Mean PPG divides the season total by the team's remaining scheduled games, so a bye week is never counted against a player, while PPG played divides by the games the player is expected to actually play.