COURT GUIDE

A projection is a promise; a rater is a receipt

Two questions come up in every category league from November on. Is the player I drafted actually doing what I drafted him for? And is the player nobody drafted doing something the rankings never saw? A projection cannot answer either, because a projection is the same number it was in October, adjusted a little. The box scores can, and a player rater is the tool that reads them: every player’s per-game line from the games he has actually played, turned into a z-score per category against everyone else’s, summed over the categories your league counts. The host platforms publish a version of this inside the app; what they do not give you is the window, the punt and the weight, and those are where the answer lives.

This page is that tool, on the same categories, punts and weights as our rankings board, so the two pages can be read side by side.

Read the rest

The window is the point

A season-long rating tells you who has been good; a seven-day rating tells you who is good right now, and in fantasy basketball those are different lists more often than a football manager would expect. A starter goes down and his backup posts a week that looks like a top-fifty season; a star comes back from injury on a minutes restriction and his last fourteen days look like a bench player’s; a rookie who was unplayable in November is a starter by February. The four windows here are the whole season, the last 30 days, the last 14 and the last 7, each ending today while the season runs. Each has its own floor, five games over a season, three in 30 days, two in 14 and one in 7, and the pool a z-score is measured against is everyone who clears the floor in that window, so a 7-day board is a comparison among the players who played this week and not against the season’s averages. One big night is a one-game window’s whole answer, which is why the games column sits beside the name: read it before the value.

How a z-score is made, and the two percentages

For each category, take every player in the window’s pool, average their per-game figure, measure the spread, and express each player as the number of spreads he sits above or below the average. Zero is the average player in the pool, plus two is elite, minus two is a real cost. The value at the end of the row is the sum across the categories in play, each multiplied by its weight, with a punted category left out. Turnovers are turned so that fewer is better. The two percentages are not scored on the percentage itself, because a percentage on its own says nothing about how much it moves a lineup: a 60 percent shooter on three attempts and a 60 percent shooter on twenty are not the same player in your weekly FG%. Each is scored on his impact, makes minus attempts times the pool’s own percentage, and the z is taken on that, which is how the category reference sites do it and how your league’s weekly percentage is actually computed. The heat on each cell is the z itself; the small figure under it is the raw per-game number.

Punts and weights, from the record

The picker is the rankings board’s. Punt free-throw percentage and the board re-sorts without it, which is how a manager punting the category checks whether the big man he took in the fourth round is actually delivering the blocks and boards that justified the punt. Weight assists at two, or blocks at three, and the board counts them that way, the way the category reference sites let a manager tilt a board toward the categories a week is decided on. Both act on the record rather than on a projection, and that is the difference between this page and the board: a punt on the board tells you whom to draft; a punt here tells you whether the draft is working. The tier ribbons between the rows mark the cliffs, and each says the size of the drop, so a tier is a fact about the gap and not a label.

Points leagues

Under the points toggle the board scores each player’s per-game line by the eleven-stat table, makes and attempts counted as pluses and minuses, and sorts on points per game over the window. There is no z here and no punt, because a points league has one number, but the window still is the point: a points manager choosing between two players for a roster spot wants the last fourteen days, not the draft-day projection.

Before the season, and what the page cannot know

Until opening night there is no season in progress to rate, so the page rates last season’s games, says so in the band, and labels every window with the season it comes from: the last 30 days of last season are the last 30 days before its final game, which is a strange window to read and is offered because it is the truth rather than a blank. The page does not know why a player’s window looks the way it does. A hot fourteen days on a team with three starters out is a fact about those fourteen days and not a forecast of the next; a cold thirty days that began with a return from injury is the same. Cross-read with the next man up page, which says whose absence is behind a lift in minutes, and with the projected board, which says what the model still expects. A player ruled out is rated on the games he played and flagged; a day-to-day player likewise. The window, the format, the categories, the punts, the weights and the position are written into the address, so the link you copy opens the same board.

The game logs are rebuilt nightly from the box scores; the roster percentages and the injury report are refreshed on our own feeds through the day. If the feed is unavailable the page says so rather than showing a stale board.