How the model thinks.
Everything below is what actually happens behind your weekly briefing — what we project, what feeds the model, what we measure ourselves on, and where we know the model is weakest.
What we project
- xPts per player per GW — expected fantasy points for the next gameweek and the horizon ahead.
- Captain & vice-captain ranking — top picks weighted for ceiling vs. floor.
- Transfer net edge — point gain of every realistic in/out swap, adjusted for the −4 hit.
- Optimal lineup — best legal XI from your squad given projections and minutes risk.
- Chip sequence ranking — when to play Wildcard, Free Hit, Bench Boost, Triple Captain across the remaining season.
Data sources
- FPL public API — live squad, prices, status, ownership, fixtures, history. Refreshed continuously during gameweeks.
- Understat — expected goals, expected assists, shot quality history. Used as model input only; never displayed raw.
Model overview
The point projection is an ensemble — multiple Fantasy-specific models trained on per-player, per-fixture features, blended into one number per player per gameweek. The lineup and transfer logic then use that projection inside a constrained optimizer (budget, position counts, 3-per-club rule, free transfers, hit cost).
Feature families the model reads:
- Form — short-window points, xG/xA trends, attacking output trajectory.
- Fixtures — opponent strength, home/away, fixture difficulty over the horizon.
- Minutes certainty — start probability given rotation history, status flags, and injury news.
- Opponent strength — defensive/attacking ratings adjusted for opponent form.
- Set-piece role — penalty taker, free kicks, corners.
- CBIT & defensive contributions — clearances, blocks, interceptions, tackles for defender bonus modeling.
Backtest summary
| Metric | Sample | Value |
|---|---|---|
| Captain pick edge vs FPL median | Season-to-date | backfill |
| xPts within 2 pts hit rate | All starting outfielders | backfill |
| Transfer net edge accuracy | Top-3 suggested transfers | backfill |
| Chip sequence accuracy | WC / FH / BB / TC | backfill |
Limitations — what the model is bad at
- Early-season uncertainty. Minutes data and xG history are thin in GW1–GW5 — projections widen.
- Post-injury return weeks. Rotation risk after a player returns from injury is consistently underweighted.
- Cup competition rotation. When clubs prioritise other competitions, lineup rotation is hard to predict from FPL data alone.
- Manager press conferences. We don't parse press-conference quotes — late team-news shifts won't reflect until the lineup posts.
Chip strategy
The chip sequencer ranks Wildcard, Free Hit, Bench Boost, and Triple Captain windows across the remaining gameweeks. It looks at fixture density (single, double, blank gameweeks), your current squad cost vs. an unconstrained optimum, expected captain ceiling for triple captain, and bench projected points for bench boost. The output is a recommended week for each remaining chip, with an honest read of how confident we are.
Versioning
Model v0.4·last updated 2026-04-29
Changelogs and methodology revisions land here as the model improves. Questions? Read the about page.