FPL AnalyzerExpected-points distribution
V8 forecasts · players grouped by projected points
Inside the models.
Beyond the prediction.
Explore the models behind FPL Analyzer. Compare what they calculate, how they run, and the evidence available for each.
Every forecast is made as soon as the official FPL site marks the previous Gameweek finished, locked before the deadline, then scored against official Fantasy Premier League results — third-party data FPL Analyzer does not control. The scoring itself is run and published by FPL Analyzer.
Explore model benchmarksLoading benchmark snapshot…
Latest scored accuracy
Each forecast is archived before the Gameweek deadline, then compared with the official FPL result for that Gameweek once it has finished. Results come from official Fantasy Premier League data (fantasy.premierleague.com), a third-party source. Forecasts are generated, archived and scored by FPL Analyzer about its own models. This is not an independent third-party evaluation.
Scored on completed Gameweeks 5, 6, 7, 8.
| Model | Category | Accuracy | Counts as correct | Error | Forecasts |
|---|---|---|---|---|---|
| xMins Model | Minutes | 87.3% | correct 60+ minute calls | Brier 0.097 · MAE 0.16 · RMSE 0.31 | 2661 |
| V8 Projections | Points | 81.6% | within ±2 pts | MAE 1.10 · RMSE 2.09 | 2661 |
| xPts Model · normal FDR | Points | 81.6% | within ±2 pts | MAE 1.10 · RMSE 2.09 | 2661 |
| xPts Model · Position FDR | Points | 81.5% | within ±2 pts | MAE 1.12 · RMSE 2.10 | 2002 |
| BPS Model | Bonus | 47.6% | within ±5 BPS | MAE 6.60 · RMSE 9.58 | 2661 |
Also available as plain text and JSON.
Benchmarks couldn’t load
The snapshot is temporarily unavailable. Try loading it again.
Accuracy evaluation
Evidence before
a leaderboard.
Pre-deadline forecasts are matched to official completed-Gameweek outcomes as soon as the scheduled evaluator runs.
- Accuracy %
- Awaiting results
- Point error MAE
- Awaiting results
- Large misses RMSE
- Awaiting results
- Probability calibration
- Awaiting results
Model explorer
One place to understand every model, component and shared interface.
| Compare | Role | Valid outputs |
|---|
Batch time is the median of seven local runs. Full projections and individual components perform different work; speed is not an accuracy ranking. Shared interfaces are not counted twice.
Select up to 3 models to compare
Model comparison
How to read
these benchmarks.
Published by FPL Analyzer about its own models. The results they are scored against are official Fantasy Premier League data, a third-party source; the measurements themselves are not an independent third-party evaluation.
Download the full snapshotExecution measurement
What counts as an accuracy result?
Predictions are recorded before the target Gameweek deadline, then matched to official live results from that same Gameweek and season. The archive records model version, player ID, input hash, timestamp, sample size and metric.
For points and BPS, the page reports MAE and RMSE in their original units. For appearance probabilities, it reports Brier score alongside MAE. Lower error is better; the page also shows a simple mean baseline for context.
Why are V6 and V8 not ranked separately?
The current V6 projection interface calls the probabilistic engine labeled V8. The xPts interface consumes those projections too. They are listed so you can trace product names to the underlying model, but do not represent three independent evaluations. xPts is archived twice — with the official FPL fixture difficulty and with Position FDR switched on — so the effect of Position FDR on accuracy can be compared directly.
Data freshness and reproducibility
The scheduled job archives forecasts before an upcoming deadline and evaluates them after official results are published. The page reads the latest aggregate evaluation and does not use personal manager or squad data.