How the numbers are made
Every number on this site comes from the same pipeline. Nothing is our opinion of a player; it is what the market, with its margin stripped out, implies. The full specification with every assumption is in the open on GitHub as docs/METHODOLOGY.md; this is the plain-English version.
1. Where the probabilities come from
We read prices for Premier League matches from an odds data feed covering dozens of bookmakers: the match result, total goals, team totals, and for each player the anytime goalscorer and assist markets. Player markets are thinner than match markets, so we show how many bookmakers sit behind every number. We keep every price we have ever read, with its timestamp, and we never overwrite history.
2. Stripping the margin
A bookmaker's prices for one market add up to more than 100%: that gap is their profit. We remove it per bookmaker, per market, before averaging. For yes/no markets with only a “yes” price (will this player score?) we take off the same margin the bookmaker builds into that match's result market. This is the biggest single assumption in the model and it is listed in the track record as such.
3. Averaging the bookmakers
Sharper bookmakers get more weight. Any bookmaker more than 12 percentage points from the rest is dropped as stale. A selection with no bookmakers left is shown as no market. We never fill the gap with a guess.
4. Who plays
There is no “to start” market on our feed, so minutes are the one input that is not a price. We combine FPL's own availability flag with the share of recent matches the player appeared in, and the share of those appearances that reached 60 minutes. Every page that shows a start percentage says so.
5. Team goals and clean sheets
From the result and total-goals markets we fit how many goals each side is expected to score, which gives the clean sheet probability (the chance the opponent scores none) and the expected goals conceded. Where a bookmaker offers a “team to score under 0.5” line we use that directly.
6. Turning it into points
FPL's scoring for the season, read from the game itself, is applied to the probabilities: points per goal by position, 3 per assist, clean sheet points when the player is expected to play an hour, deductions for goals conceded. Three components are estimates from the player's own season so far rather than market prices, because no market exists for them: saves, bonus, and defensive contribution. They are labelled as estimates wherever they appear.
7. Captaincy, differentials and the rest
Captain candidates are ranked by expected points doubled. Blank risk is the chance of neither a goal nor an assist. Differentials compare the market's goal probability with how many managers own the player. Fixture difficulty blends win probability with expected goals conceded.
8. Why a number moved
“Prices moved” shows selections whose probability changed by 4 points or more (5 for clean sheets) between our previous refresh and the latest one. The market moves when team news lands, when a manager speaks, or when sharp money arrives; we show the move, not a story about it.
9. Bonus points
Bonus goes to the players with the most bonus-point-system points in a match, which depends heavily on whether the team won and kept a clean sheet. We keep a table, from every match this season, of average bonus by position, home or away, result and clean sheet. A player's projected bonus is that table weighted by the market's result and clean-sheet probabilities, scaled by the player's own bonus record shrunk towards the average for the position. It replaces the flat season rate wherever a match market exists.
10. Chips and the weeks ahead
Doubles and blanks are read from the fixture list, so they are known as soon as the league publishes them. Points for a later gameweek exist only where the bookmakers already price the match: we take the team's expected goals and clean sheet from the match market and give each player his season share of the team's goals and assists. That is labelled “team markets”; weeks with no market say so. Bench Boost value is the bench points that week, Triple Captain the best captain's points once more, Free Hit the gap between the best £100.0m squad the solver can buy and your own best eleven.
11. Price changes
Prices move once a night from net transfers and the game does not publish the rule. We count each player's net transfers since 2am UK as a share of his current owners and call a move when it passes a threshold. The threshold starts as a placeholder and is re-learned from our own hits and misses once twenty moves have been seen, so the page carries an “experimental” label and its running record until the self-scoring says it is reliable. This is the one feature on the site with no market behind it.
12. The top 10k and your mini-league
An hour after each deadline we read 1,000 public teams spread across the top 10,000 ranks of the overall league and count who owns, starts and captains each player. Effective ownership counts a captain twice. For a mini-league we read each rival's last squad the same way and project it on this week's numbers. We only ever read public pages.
13. The release rule
At every deadline we freeze the projections. When the gameweek is settled we score them: error in expected points, whether our captain pick beat the crowd's, how well the clean sheet and minutes probabilities were calibrated, and whether the market's XI beat the template. A feature only goes live after two consecutive gameweeks of correct output, and the scores stay public on the track record whether they are good or bad.
14. What we will never do
Invent a number when the market has none. Show a price instead of a probability. Overwrite history. Send you anywhere other than fantasy tools.