Who is actually going to play

Nobody reaches a hundred per cent and nobody should. The model anchors every player against one round at even odds, so a player who has started every match of the window comes out in the low nineties — which is about what a nailed-on starter actually manages once rotation, knocks and suspensions are counted.

The expected minutes beside the band are the whole of the estimate: his chance of starting times what his own starts have been worth, plus his chance of coming on times what a substitute appearance is worth. It is not a prediction of one afternoon.

A short history is labelled rather than left out. Under 2 stored rounds the estimate is resting on his season totals and the model's own anchor rather than on rounds anybody can see, and the row says so.

The news column is the game's own, unedited, and it is the one input here no statistics feed carries. It arrives with the feed, so a suspension or an injury announced since the last refresh is not on this page yet — the refresh time is at the top.

A player with excellent underlying numbers and a poor band is shown as exactly that, and is neither hidden nor quietly demoted. The projection beside him already has his minutes in it; the band is there so the reader can see how much of that projection is a bet on selection.

The same band is the minimum-security filter on every board in this section, read off the same figure — so a board narrowed to “Safe” and this page's own safe column can never disagree about a player.

Every ranking here is in expected points or in the weighted score for the rounds ahead, which are the units this model is measured in. The value column is those figures restated over a price — it is a way of scanning the board, and nothing on this site backtests it.

The board is every player the game would let you field, the flagged included — whether a doubtful player is worth picking is a judgement the projection makes with his published chance of playing, not one made by hiding him.