2025-26 · GW 27–36 · model v6-phase9

Our captain landed in the top 5 on 0 of 10 gameweeks

We've captured 25.6% of the perfect-captain points pool this season.

CoachMe captain

25.6%

Share of optimal captain points we captured.

Always-Haaland baseline

5.9%

What a “captain Haaland every week” strategy would have captured.

We're beating the naive baseline by 19.7%.

Transfer net

-0.60

Average points the player we moved IN beat the player we moved OUT by.

Hit take rate

20.0%

How often a recommended -4 hit produced net-positive points.

Bench order

1.70 / 3

Bench positions ordered correctly, averaged across GWs.

Per gameweek

Our captain points vs the optimal captain points each GW.

Calibration

When we predict a bucket of expected points, what do players in that bucket actually score on average? The line should hug the diagonal.

Are our predictions trustworthy?

Per-player xP is structurally flat — the ensemble shrinks toward the mean. We publish the three metrics below so you can see exactly how spread out our predictions are, how often actuals stay under our ceiling, and whether each predicted-xP bucket pays off.

Variance ratio

0.59

std(predicted) / std(actual). 1.00 = equally spread; we are more compressed than reality.

n = 7,826 player-GWs (≥5 GW history)

Actual ≤ ceiling_xp

86.4%

How often actual points landed under our 80th-percentile ceiling. Expected ~80% if well calibrated.

n = 7,838 player-GWs

Calibration by predicted xP bucket

For each predicted-xP bucket, mean actual points. If we are well calibrated, the “actual” bars track the dashed y=x line.

What we got wrong

The three gameweeks where our captain call missed the top scorer by the most. No spin — here they are.

GW 28

We captained Wirtz. They scored 0.

Top score: 18

missed by 18 pts

GW 32

We captained B.Fernandes. They scored 4.

Top score: 21

missed by 17 pts

GW 33

We captained Enzo. They scored 4.

Top score: 20

missed by 16 pts

Want these picks delivered every gameweek?

Last updated 14 May 2026, 12:08