About FieldTilt

The numbers behind the numbers.

What is FieldTilt

FieldTilt is a Premier League form model for FPL managers and fans. It surfaces the form trends, breakout players, and value picks that most stat sites bury, and it's built to spot players heating up before the crowd notices. All in one dashboard, updated daily. And unlike most tools, it shows its working: the full backtest, wins and losses, is right here on this page.

Tilt Score Methodology

The Tilt Score is a proprietary rating from 0 to 100 that captures a player's current form by blending two dimensions: per-90 efficiency (how productive a player is per minute on the pitch) and total output (rewarding sustained excellence over many games).

Blended Approach

Rate and volume are weighted and log-scaled to compress outliers, keeping the 0–100 range meaningful. One big haul won't inflate a score the way consistent returns will.

Recency Matters

An exponential decay weights recent gameweeks more heavily, so a goal scored last weekend counts for more than one scored in September. Recent form shapes the score most.

Position-Aware

Forwards are rated heaviest on goals, midfielders balance goals and assists, while defenders and goalkeepers are rewarded for clean sheets. Each position is judged by what matters most.

Confidence Filter

Players with limited minutes are scaled down proportionally. This filters out noise from one or two strong cameos and ensures only players with a meaningful sample size rank highly.

The 6 Card Stats

Every player card shows six stats rated 0-99, each distilled from underlying Premier League data:

ATTAttacking output: goals scored blended with expected goals (xG).
CRECreativity: assists, expected assists (xA), and FPL creativity index.
DEFDefensive contribution: tackles, recoveries, clearances & interceptions.
FITPhysical availability: minutes per start and start rate (durability).
FRMForm: maps directly from the Tilt Score. Current momentum in one number.
VALValue: FPL points per million spent. Finds the bargains others miss.

Does it actually work? We show our working

A form score is only worth anything if it predicts what happens next. Most “AI picks” tools never show you a test they lost. We will, because the honest version is the whole point.

First, the humbling part

We backtested the Tilt Score point-in-time across four Premier League seasons (no hindsight: at each gameweek it only sees data available at the time) and asked one question. Does it rank players for their next 3 gameweeks better than the free, obvious alternatives? Correlation with what actually happened (higher is better):

Total points to date(the dumbest baseline)0.415
Tilt Score (v1)(our original score)0.390
Points per game0.387
Last 3 GW points (pure form)(the worst of all)0.313

Our score lost to simply sorting by total points. And recent form, the thing the score leans on, was the worst predictor of the lot. We could have buried that. Instead we dug into where it genuinely had an edge.

Where it actually wins: timing

Total points just points you at the players everyone already owns. The edge is spotting the next one early, and there the form signal is far stronger. Predicting who rises in price over the next 3 GWs:

Predicting price rises

~3× better than total points

Tilt 0.114 vs total points 0.038. It catches risers before the crowd.

Predicting ownership growth

Positive where total points goes negative

The big names are already owned, so they can't grow. Form finds the ones about to.

So we rebuilt it

We added what the score was missing: underlying process (expected goals and assists vs actual output). Then we fit the new version out of sample, trained on two seasons, tested on a third it had never seen. Predicting next-3-GW points on that held-out season:

Total points (the baseline to beat)0.428
Old score (v1)0.406
New score (v2)(finally beats it)0.443

Honest caveat: the improvement is modest, about 3%. Fantasy football is largely efficient; anyone selling you a magic edge is selling you something. What we have is a small, real, measurable one. It's better at spotting risers early than the obvious baselines, and it now shows its working, including the version that embarrassed us.

Where this stands today: the rebuilt engine is validated and in final testing, and goes live for the 2026/27 season opener (GW1). Until then, the Tilt Score you see across the site is our current model, and this page will always show the latest honest numbers.

Want to see it play out rather than take our word for it? Every captain we would have picked over the run-in of last season, hauls and blanks both, is on our track record. No cherry-picking.

Data Sources

  • Official Premier League stats via football-data.org (Opta), goals, assists, cards, and match results.
  • FPL API, player metadata, clean sheets, form data, gameweek history, xG, xA, creativity index, and ICT data.
  • Updated daily with hourly checks during active match windows.

Get in Touch

Got feedback, a feature request, or a data question? Reach out at hello@fieldtilt.com

Disclaimer

FieldTilt is an independent project and is not affiliated with the Premier League, Fantasy Premier League, or any football club. All trademarks belong to their respective owners. Stats are provided for informational purposes only and should not be taken as betting advice.