Methodology

Three sources, all public or built from public data.

• FPL's public data. Players, prices, ownership, fixtures with difficulty, match statistics (including expected goals and assists), availability news, the rules of the game, live scores and every manager's public team. Pages refresh it every one to fifteen minutes depending on how fast it moves.
• The analytics warehouse. A daily snapshot of FPL's data, modelled into tables the public data cannot answer on its own: team attack and defence by expected goals, trends, consistency, set-piece takers, availability history and ownership among the best managers.
• The projections archive. The xPoints feed and the site's own projections, saved at every deadline and never edited, with the grades written beside them after each gameweek.

The site stores nothing about you. A Team ID is kept in your browser and used to read your public team from FPL.

xPoints is one number per player: the points expected in the next gameweek. Two sources can supply it. The baseline is FPL's own expected points. The challenger is this site's model, which runs as a shadow model: it produces numbers for every gameweek, they are frozen at the same deadline, and they are graded against the same results, but they are not shown as the feed.

The challenger takes over only through the promotion gate: over the last six gameweeks in which both were frozen, it has to beat the baseline at ordering the players who started, by a margin that is clear of chance. Until then the feed is FPL's figure and every table says so, with the gameweek it targets and the time it was generated. The current state of the gate, with the reasons when it has not passed, is on the Accuracy page.

Whichever source holds it, the feed is only used when it names the next gameweek and this season and carries the time it was made. Past three days old it is marked stale, and past a week it is refused and the tools say so instead of planning on it.

The reason for the caution is that a baseline resting on two games of form is easy to beat in one week by luck. Six paired gameweeks with an interval is the least that separates a better model from a lucky one.

Every projection is frozen at the deadline and scored after the gameweek against FPL's official points. Three measures, because each catches a different failure:

• Average miss (MAE), over every player, set beside the score for predicting zero for everyone. Most players do not play, so a feed has to beat that to be saying anything.
• Rank match, among players who started, with a 95% range. Choosing between players is a question of order, so this is the measure the promotion gate uses.
• Top 20: how many of the feed's top 20 finished in the real top 20.

The site's multi-week projections are graded the same way by how many deadlines ahead they were made, and the chance of sixty minutes is graded with a Brier score and a calibration table: of the players given 70%, did about 70% play sixty?

The feed covers one gameweek. The planner, the chip advisor and the value tables need several, so the next gameweek's number is taken apart and rebuilt for each gameweek over the horizon, fixture by fixture:

points = scoring rate x minutes factor x opponent factor x availability gate

• Scoring rate. The feed's number with next week's minutes and opponent divided out: points per fixture at full minutes against an average side. It is capped, and without a usable feed number it falls back to the season's points per game.
• Minutes factor. The share of ninety minutes that xMins expects per fixture.
• Opponent factor. From the opponent's attack or defence by expected goals, with home games at 1.05 and away games at 0.95. Only half of the opponent's strength is credited, since the feed already reflects the next fixture in part.
• Availability gate. FPL's chance of playing applies to the next gameweek only. Later weeks read the return date in FPL's news; a player out with no date is out for the whole horizon, and a loan clause zeroes the gameweek it names.

A blank scores nothing and a double counts both games. Each further week ahead leans a quarter less on the feed and more on what players of that price and position usually score, so one fluke number does not become a season.

xMins is deliberately small. It looks at a player's last six gameweeks, weighting the most recent fully and the oldest at a fifth, and estimates four things: how often he starts, how long he plays when he does, how often he comes on when he does not, and for how long. A mild prior (an even chance of starting) keeps the first weeks of a season honest.

FPL's availability multiplies the result, fixtures in the gameweek are added up, and the outputs are the expected minutes, P(start), P(60+) and a role.

The chance of sixty minutes is a blend: the site's own estimate averaged with the xPoints model's, which reads the transfer market as well as recent starts. The two use the same definition, so they are graded side by side, and the blend was adopted after the model's half graded better on the same players. Player pages show both halves; minutes, the chance of starting and the role stay the site's own.

It knows nothing about press conferences, tactics or a manager's mood. It is a projection of a pattern, not a team sheet.

FPL's form ignores who the points came against. Fixture-adjusted form takes the last five gameweeks and scales each game's points above the appearance points by the opponent: the defence faced (expected goals conceded per game) for midfielders and forwards, the attack faced (expected goals per game) for keepers and defenders.

Team strength is shrunk towards the league average while few games have been played, and the multiplier is capped between 0.6 and 1.5 so one lopsided week cannot dominate. Player pages show the working game by game.

PAR asks what a player adds over the option anyone can afford. Replacement level is the median of the best eight playing options at or below the floor price for the position (4.5m keepers and defenders, 5.5m midfielders, 6.0m forwards), where playing means sixty expected minutes next gameweek. A second level compares him with players within half a million of his price.

PAR per million divides by the price above the floor. It is worked out on the multi-week projections when they exist and on the season's points per game otherwise, and each player gets a verdict and a rank within his position.

The captain decision has two halves. The first is points: a simulation of each candidate's gameweek, several thousand times, from his own rates. Goals and assists come from xG and xA per 90, with bonus, cards, defensive contributions and his team's expected goals conceded; this season is blended with last by minutes played, and thin samples are pulled towards a typical player in the position. The result is a mean and a shape: the chance of a haul, a return and a blank, a floor and a ceiling.

The second is the field. Effective ownership decides whether a captain protects your rank or attacks it. FPL does not publish captaincy before the deadline, so the share is estimated from ownership and expected points across the leading candidates and labelled as an estimate. The picker then names a safe captain, a differential when one is close enough on points, and a vice who plays no earlier than the captain.

Both start from your public team: your squad, your bank, your selling prices from your transfer history, and your free transfers worked out from your gameweek history. The squad limits, the budget, the banked-transfer cap and the sale terms are read from FPL's published rules rather than assumed.

The objective is the honest one: the expected points of the best legal eleven after the transfers, captain doubled, minus the best eleven before, minus any hit. A swap only counts for what it adds to the team that actually plays.

The one-week optimiser checks every one and two transfer combination among the best and cheapest candidates in each position, and builds three-transfer plans on the best pairs. The Transfer Planner searches three to eight gameweeks ahead, keeping the most promising paths each week, with later weeks discounted because they are less certain. It banks transfers correctly, lets you lock or exclude players and aim for a target, and reports the gain against simply rolling.

The Chip Advisor puts every remaining gameweek on one scale, so that weeks can be compared. For each chip it works out what the chip would add in each gameweek it can still be played: the bench's points for Bench Boost, the captain's for Triple Captain, the best squad your budget can buy against your own for Free Hit and Wildcard. It then sets each week against a typical week, because a chip is worth playing when it beats the alternative weeks, not merely when it is positive. Doubles and blanks that FPL has not scheduled yet cannot be seen, and the page says which fixtures are still to be placed.

FPL now publishes its own price-change tracker for every player: progress towards the next change, the current hourly rate, its own projection for the next three nights, and whether the player is locked. Price Predictions reads those fields directly and adds the net transfers still needed. It replaced an older estimate from transfer counts, which could not know how far along FPL's own accumulator each player was. Early in a season FPL flags the tracker as calibrating, and the page repeats that warning.

FPL's ownership counts every team, including abandoned ones. Top 10k and top 100k ownership comes from a sample of managers in each group, read from their public teams each gameweek, with the sample size shown. Inside your own mini-league nothing is estimated: ownership and effective ownership are counted from the managers' actual picks.

• Football is low scoring and FPL points are lumpy. A good projection is right about order more often than not; it is wrong about any one player most weeks.
• Team news after the numbers were made is not in them. Check the day's news before the deadline.
• The first gameweeks of a season rest on very little. Tables that need a run of games say when they unlock.
• Multi-week numbers are less certain the further ahead they look, which is why the planner discounts them.
• Fixtures that FPL has not scheduled, rotation in cup weeks and transfers between clubs are invisible until they appear in FPL's data.

Every number is a projection. None is a promise, and nothing here is betting advice.

When a model changes, this page is updated with the date, and the Accuracy page goes on grading every gameweek, so an improvement has to show up in the record. When something is wrong, tell us through the contact page; corrections are made in place and dated. Frozen projections are never edited, whatever happens afterwards.

The site is independent and is not affiliated with the Premier League or Fantasy Premier League.