FPL is 85% skill, measured not argued

Ask a room of Fantasy Premier League managers whether the game is skill or luck and you will start an argument that outlasts the season. Someone finally measured it instead: 85% skill, across 24,041 complete manager seasons.

That number comes from Zakariae Boui's fpl-strategy-mcp, an open-source, rules-aware FPL decision engine built with unusual discipline. The methodology is walk-forward with no hindsight: trained on 2016/17 through 2023/24, selected on 2024/25, and tested on 2025/26 held out entirely. His patient-chips candidate scored 2,486 points on that untouched season against a 2,413 target, good for an estimated rank of 2,324 out of 13.1 million, with zero hits taken. The benchmark also profiled elite play: the top-100 median manager gained 325 points net from transfers, made 62 transfers, took 4 hits, and agreed on the captain pick 71.1% of the time.

Here is the paradox that makes the study worth your time: the champion wins a replay only 3.6% of the time. Skill decides your season; luck decides the trophy. Over 38 gameweeks, process compounds and the best decision-makers rise. In any single week, or in the title race itself, variance runs the show.

The edge lives in process, not outcomes

That split should sound familiar. It is the same shape as everything expected goals tells us: price the process, not the outcome. A 1.28 xG attack that scores five is not a five-goal attack, and a manager who wins a gameweek on a differential captain is not necessarily a better manager. What compounds is the repeatable part: transfers with positive expected value, captaincy on the highest-projection player, chips played with a plan instead of a panic.

The study's elite benchmarks read like a checklist for exactly that. Sixty-two transfers across a season is active but not chaotic. Four hits says patience beats tinkering. A 71.1% captain agreement says the best managers mostly agree on the obvious pick and win on the margins around it. None of this requires genius. It requires a process and the discipline to run it 38 times.

What to do with this

If you play FPL, the takeaway is unglamorous and valuable: build the process, then let the season do the work. Judge your gameweek 12 captaincy by whether it was the right call with the information available, not by whether he blanked. The managers at 85% skill are not predicting the future; they are making slightly better decisions slightly more often, and letting arithmetic handle the rest.

And if you want the same discipline applied to the underlying football: the xG feeds, league simulations, and edge alerts behind Dribble's weekly recaps are process inputs of exactly this kind. Skill compounds. Luck does not, which is why you price the process.

Want the underlying data? The same xG feeds behind Dribble's recaps are available as BigQuery tables, API endpoints, and downloadable files. See plans and pricing →

Methodology note: figures are from an independent open-source study (fpl-strategy-mcp, MIT), not from Dribble data. It is one builder's project, not peer-reviewed research, but its walk-forward, no-hindsight discipline is the standard serious backtests are held to.