How to read the tables

The Offense and Defense pages are one row per team and one column per position. A cell is fantasy points per game under the scoring system in the control bar: on Offense the points that team's players at the position produced, on Defense the points opponents at the position scored against it.

Polarity colours

Each column is coloured on its own range, red through white to green. Green is the end that is good for the row: on Offense a team producing a lot at a position, on Defense a team allowing a lot, which is what a matchup wants to find. Columns with no polarity (salary, games played) stay white.

Ranks

The small grey superscript is the team's rank in that column, 1 being the highest value. On the Defense page rank 1 allows the most fantasy points at the position.

Prior weight superscript

An italic amber superscript such as 44% means the prior carries that share of the cell. Early in a season a cell is (n × observed + k × prior) / (n + k), so with four games played and k = 4 the prior is half the figure; the tooltip on the cell spells out n, k, the observed figure and the prior. The badge appears once the prior weight passes 25%.

Sorting and drilling down

Click a column header (or focus it and press Enter) to sort; click a cell to open the games behind it. The first column links to the team page.

The Window control chooses which games count: the whole season, the last 5 to 15, or a weighted window where each game's weight halves every half-life games (NFL 3, NBA and NHL 6, MLB 15). A weighted window reports its effective sample n_eff rather than the raw count.

Every cell is shrunk toward a prior with weight k in game-equivalents (NFL 4, NBA and NHL 10, MLB 25). The prior is last season's figure for the same team and position, blended with the league average by the team's continuity (how much of last season's usage is still on the roster, less any coaching-change penalty), or the league average alone when no previous season is loaded.

Solving for x. While no team has played more than a few games (NFL 4, NBA and NHL 15, MLB 25) the Offense and Defense pages switch to a view that lists the prior, the observed figure and their difference per position, sorted by how far a team has moved from what last season predicted. The banner above the table says so and shows the prior weight; the checkbox in the control bar turns it off.

A cohort is a tag on games: the starting quarterback out, the top scorer out, the starting goalie out, a dome, heavy wind, a neutral site, overtime, a back-to-back. The Cohort control filters every cell to the games where the tag applied; with Compare on, the table shows all games and the cohort side by side with the difference.

Cohort cells shrink toward the team's overall figure with twice the usual k, so a cohort of two games never reads as a trend. Tags are derived from the box scores when a game is scored and re-derived whenever a season is rescored.

The Matchups page puts a team's own points for at each position beside tonight's opponent's points allowed at that position. The projection is the geometric middle of the two relative to the league average, shrunk as above, then multiplied by the Vegas factor: the game's implied team total against the team's average points scored, capped between 0.7 and 1.3.

A player's forecast is his share of the team's position total times the opponent's allowance, so an injury that removes a starter redistributes the share to the players left. Floor, median and ceiling come from the player's own distribution of scores in the window.

Reference. Every price is measured against a reference: for game markets the vig-free DraftKings probability (both sides' implied probabilities scaled to sum to one), for props the site projection and its distribution. The model's own probability sits beside it.

Edge. Model probability minus the venue's implied probability after that venue's fees (Kalshi and Polymarket charge on winnings; a sportsbook's cost is its vig). The Opportunities page lists edges across venues, largest first, and only after costs.

CLV. Closing line value is the difference between the price a pick was recorded at and the venue's close, in probability points. It is graded after the final alongside the result; the public record reports both, and shows nothing but a count until twenty picks have been graded.

The Player Pool lists every player on a slate with salary, floor, median, ceiling, boom rate, the Vegas multiplier, injury status and confirmed lineups. The Lineup Builder takes that pool and an objective (cash: the floor; GPP: the ceiling with correlation), honours locks and exclusions, caps exposure across a set of lineups, stacks a passer with his receivers or a game's two offences, and exports the site's CSV.

Published lineup sets are scored against the box scores once the slate's games are final; the DFS track record on the Lineups page keeps every set, its actual score and where it would have placed in uploaded contest standings.

Alert rules fire after every odds, props and model sweep: a key-number cross, a line move, a prop move, a model edge, steam, books disagreeing. Each rule fires at most once a day per market and goes out by email, a daily email digest, SMS, or browser push once a browser is subscribed on the Alerts page. Quiet hours hold emails for the digest.

The ledger records picks with their price and venue; every pick is graded after the final with its result and closing line value, and the CSV export carries every column.