BEAR BONESFPL · BY BEARBET Your team
Every gameweek, good or bad

How accurate are our FPL projections?

Our projections have been closer to the real points than FPL's own in 8 of 8 gameweeks we've graded, measured by the average miss per player who played. The team our model locks before each deadline has matched or beaten the gameweek average in 4 of 8. Both can be true at once, because a single gameweek is mostly luck. Nothing here is edited after the fact.

Predicted is the locked total of the team's projections. Teams picked from the best projections have scored about 5.5 points below that total, last season and this one, because the pick also collects the players the model is most optimistic about. The score we quote for the next team takes that off.

Average miss per player, ours against FPL's (lower is better)

GameweekOursFPLTop 15 found (ours)Top 15 found (FPL)
GW31.842.4321
GW42.142.2922
GW52.062.1812
GW61.912.2711
GW71.932.4932
GW81.992.5720
GW91.741.9045
GW101.912.1332

The model's locked team

GameweekCaptainPredictedScoredAverage
GW3Haaland67.54351
GW4Palmer66.46769
GW5Haaland68.14748
GW6B.Fernandes69.34258
GW7Haaland62.78159
GW8Haaland59.75545
GW9Haaland57.16447
GW10Haaland62.96355

Bear is the coach!: the model's own FPL team

The bear also runs a real FPL team on this model, with transfers, hits and chips, since GW4. It has 473 points so far. From GW4 to GW10 it scored 473, which would rank about 207,000 if every team had started counting in GW4 (estimated from 1,500 random teams). See it on FPL.

If the model had started in GW1 (rebuilt afterwards)

The model made its first pick in GW3. We rebuilt GW1 and GW2 later using only what was known before each deadline: earlier form, pre-kick-off market odds and archived predicted lineups. Nobody saw these teams before kick-off, so they don't count in the record above.

GameweekCaptainPredictedScoredAverage
GW1B.Fernandes63.64850
GW2B.Fernandes64.19781

Get these calls for your own team