Everything the model produced, in the form it was produced in. The tables below are the run record flattened; nothing is recomputed on this page.
An interactive view of the same data: the squad against its prediction, all 561 candidates sortable and filterable, the per‑gameweek decomposition for any player, model calibration against what actually happened, and the snapshot timeline.
Open the explorer — a single self‑contained page of 801 KB; once it has loaded it needs no network.
CSV, comma‑separated, UTF‑8. Open them in a spreadsheet.
| File | Rows | Size | What is in it |
|---|---|---|---|
| players.csv | 561 | 79 KB | One row per selectable player: price, ownership, our expected points for each of the five gameweeks, P(start), and the actual gameweek 1 return beside the prediction. |
| components.csv | 2,805 | 370 KB | The decomposition. One row per player per gameweek with all nine terms broken out — appearance, goals, assists, clean sheet, defensive contribution, saves, goals conceded, cards, shrink. |
| history.csv | 3,670 | 246 KB | Every field that moved between consecutive snapshots: price, ownership, status, news, chance of playing. This is the one table that cannot be rebuilt after the fact. |
| squad.csv | 15 | 1 KB | The fifteen actually entered, predicted against actual. |
Built from snapshot b7f5cf2940cb5b25, taken 2026-08-21T09:45:01Z, under formula v1.0-phase0 and rules 2026-27. The method is described on the method page.
The source data is the public Fantasy Premier League API, captured every fifteen minutes, plus the community match archive. This site is a personal project and is not affiliated with the Premier League or with Fantasy Premier League.