xgEdge

Fantasy Premier League

How the fantasy projections work

Every projection here is one player's expected minutes carried through his club's next match. The attacking half comes from his own per-90 rates, the defensive half from a score matrix fitted for the fixture. What follows is each part in turn, and which of them has actually been measured.

What the projections are built from

Two feeds, and the half of the scoring table each of them answers. Fantasy says who a player is, what he costs and how often he scores; this site says what his club's next match looks like.

The defensive half
A clean sheet is the probability the opponent scores exactly nothing, read off a score matrix fitted for the fixture — not one minus an average. The concession penalty is the expected number of two-goal steps, which is not the same as the step at the expected number.
The attacking half
Goals and assists come from Fantasy's own per-90 rates, rescaled to the fixture. A striker's rate is moved by how much better or worse this opponent is than the ones the rate was earned against, and clamped so one easy match cannot triple it.
Minutes
Nobody publishes team sheets in advance. The start probability is estimated from recent rounds and the game's own fitness flags, and it is the largest source of error here. That is why every player carries it as a phrase — Expected to start, Likely to start, Rotation risk, Doubtful or Unlikely to start — rather than having it folded into his total.
Start labels
The start probability is printed as a phrase: Expected to start is 90% and above, Likely to start 75–89%, Rotation risk 50–74%, Doubtful 25–49% and Unlikely to start under 25%. Every list prints the percentage beside the phrase except the Players list, which keeps it in the phrase's tooltip. The figure is the estimated probability he starts the round ahead, from his recent and season starts, capped by what the game says about his fitness. The model never says more than about 92% — a full run of starts is still pulled a little towards the average — so Expected to start means at that ceiling. A row at Rotation risk or worse whose projection is in the top 10% of its page carries a minutes flag. The Minutes page groups the phrases Expected to start, Likely to start, At risk (Rotation risk and Doubtful) and Unlikely to start. Value over replacement is a player's projected points over the window less his position's replacement level, the best projection among players at or under a cheap price line: 4.4m for goalkeepers, 4.0m for defenders, 4.5m for midfielders and 5.5m for forwards, the price where a regular starter first appears. Value is labelled Great, Good, Average or Poor at the quartiles of value over replacement within his position, taken over the players at 50% or more to start. The Defence page groups defenders and midfielders Elite all-round, Safe, Attacking and Risky by the defensive half of the round ahead and xGI per 90, each cut at the middle of his position over the players at 75% or more to start.
Decisions and their cuts
Every cut on the section is one constant. A captain call is Clear when the best is more than 0.5 projected points ahead of the alternative, and Close inside that; a comparison inside the same 0.5 points is too close to call. A differential captain must be owned by under 10% and sit within 1.0 projected points of the best. The hub's transfer target is the top of the Players board among players Expected to start or Likely to start. My Team recommends a transfer only when it is worth at least 2.0 projected points over the next five gameweeks, added up round by round with every hit of 4 taken off; a smaller best move reads Roll your transfer, a hit under that line reads Marginal, and replacing a player the game has ruled out comes first whatever it gains. Needs attention flags a starter red at Doubtful or Unlikely to start, or when the game has ruled him out; orange at Rotation risk, when his run over the next five rates Very tough on his line, or when he is the weakest starter and his best affordable replacement adds 2.0 projected points or more. It shows at most 5, red first.
How the lists are ordered
The Players page lists everyone in the order of the figure it prints beside each name: projected points over the rounds you pick, added up plainly, or the one round under GW. Level totals fall back on the ranking score, then the name. The other boards, the transfer search and the comparison rank on that ranking score instead, which counts each round 0.80 of the one before it, so the deadline in front of you weighs most and a decision made now is judged mostly on the round it affects first. On the Players page the ranking score is one sort among the others, a click away, and its figure is in each player's details.
Fixture ratings
A fixture's attack rating is the expected goals our model gives the club in it, over the club's average across every match of the season it has rates for; its defence rating is the opponent's expected goals, over the league's average, and lower is better. Attack is Great at 1.13 or more, Good from 1.03, Neutral above 0.97, Tough above 0.88 and Very tough at or under it. Defence is Great at 0.84 or under, Good to 0.93, Neutral under 1.04, Tough under 1.13 and Very tough from it. The cuts sit near the fifths of the fixtures ahead when they were set, on 30 October 2026, so each word covers about a fifth of them. The fixtures page's Best attacking outlook card and Most expected goals sort rank clubs on the goals they are expected to score over the window, not on the ratings: a weak attack with a soft run rates Great and still scores less. Its Best defensive outlook card and Fewest conceded per match sort rank clubs on the goals they are expected to concede per priced match, not in total, because a total is smaller for every match it is not over and a blank would put a club at the top for the match it does not play. A player's run is rated on his own line, attack for a midfielder or forward and defence for a keeper or defender: the mean of its fixtures' ratings, on a ladder of its own, because a mean of five sits far closer to the usual than one fixture does. Attack is Great at 1.10 or more, Good from 1.05, Neutral above 0.95, Tough above 0.90 and Very tough at or under it; defence is Great at 0.90 or under, Good to 0.95, Neutral under 1.05, Tough under 1.10 and Very tough from it — 10% and 5% either side of the usual. Best fixtures on the Players page, Buy low's fixtures gate and My Team's Poor fixtures read it.
Where the two feeds meet
A player matched to this site's own record is projected partly from expected goals measured here; an unmatched one is projected from Fantasy's figures alone. The share is stated rather than hidden, and every unmatched row is marked.
What the track record grades
Every settled round is graded once and kept on the track record: the xG Edge XI and its captain projected against what they scored, and the model's error a player against two plain baselines over the same sixty players — each player's own average so far, and FPL's own forecast frozen beside ours. Transfers, chips, the start labels and the floor, ceiling and chance of a haul are not graded there; nothing has scored them yet.

What we've tested

  • Backtested

    Projected points

    xG Edge error
    1.617
    Simple baseline
    1.842

    Average error in points per player, across 456 players. The baseline predicts each player's average so far; the test replayed gameweek 2 after it was played, recorded Sunday, 6 September 2026.

  • Graded

    Projected points, every round since

    Graded live on 8 settled rounds, each scored from forecasts frozen before its deadline: average error 2.74 points a player against 3.16 for each player's own average so far, better in 7 of 8. Against FPL's own forecast, frozen beside ours: 2.74 against 3.27, better in 7 of 8. See the track record

  • Not validated

    The floor, the ceiling and the chance of a haul

    Percentiles of a week built from the same rates as the projection. They agree with it by construction, but no floor or ceiling has been checked against a round anybody played.

  • Not validated

    Form score

    How far a player's returns run ahead of or behind the chances he had, over the rounds stored. Nothing here has checked whether a gap that wide closes.

  • Not validated

    Squad rating

    A score out of a hundred built from ten lines under weights set by hand. No rated squad has been checked against what it went on to score.

  • Not validated

    Recommendation thresholds

    The cuts behind every label and recommendation: the start labels, the margin a transfer must clear, the fixture and value labels, and the margins under which a page calls two choices too close to call. Each is one figure set by hand and stated on the How it works page; nothing has measured what following them costs.

84% of 504 players matched to this site's own data Prices, ownership and fitness flags read from Fantasy 17/11 16:00

And the model underneath it

Everything above is the layer this section adds: a scoring table, a minutes model, and two feeds rescaled into points. Under it is the expected-goal model the whole site runs on: how a match is fitted, what the ratings mean, and what the model has been measured against.

The model in force is version 1, since 26 August 2026. Each club's ratings are fitted on its latest 20 matches with expected goals, on a half-life of 107 days, with shot volume weighted 0.5 beside expected goals; a club with few matches is pulled toward average by 5 matches' worth of an average side; and each competition's baseline is its expected goals, with no correction toward the goals it scores.

How the model works →