How to pick an FPL captain with a model instead of a hunch

The armband is a risk decision

Doubling a player doubles his variance as well as his mean. That is obvious and almost universally ignored. The right captain depends entirely on what outcome you need: if you lead your mini-league by 30 points with five gameweeks left, your job is to not lose. If you trail by 60, a safe 12-point week is functionally the same as a blank.

So the model produces two shortlists rather than one. The safe list ranks by simulated floor, the 10th percentile of 500 runs. The ceiling list ranks by the 90th percentile. They frequently disagree, and the disagreement is the useful information.

Minutes certainty comes first

Before comparing distributions, filter on the probability of playing 60 minutes or more. A captain who is subbed at the hour, or rested entirely, costs you the week regardless of how good his rates are.

Anything below roughly 80 per cent to reach 60 minutes should be treated as a ceiling-only option, and below 65 per cent it is a gamble rather than a captaincy decision.

Reading the two lists

• —Both lists agree on the same player: the easy week — captain him and move on
• —Safe list favours a nailed premium, ceiling list favours a differential: your league position decides
• —Top of the ceiling list has a low floor and a low minutes probability: that is a punt, price it as one
• —Nothing clears 6.0 xPts: a blank-ish gameweek, prioritise floor and avoid clever

Effective ownership changes the maths

Captaining the player 70 per cent of your league already has doubled is a defensive move: you cannot gain, you can only avoid losing. Captaining a 5 per cent-owned player is an attacking move with a much wider swing in either direction.

That is why captaincy advice that ignores your specific mini-league is incomplete. Global ownership is irrelevant if the twelve people you actually compete with all own the same three forwards.

Head-to-head is a different problem

In an H2H league you are not maximising points, you are maximising the probability of outscoring one specific opponent. If their squad is projected 8 points above yours, the safe captain guarantees a narrow loss and the volatile captain is correct even though it lowers your expected score.

The model handles this by treating each duel as a distribution of margins and picking the armband that maximises the chance the margin lands above zero.

Measuring whether it worked

Captain efficiency — the recommended captain's actual points divided by the best possible captain's points that week — is the honest scoreboard. Nobody hits 100 per cent. Consistently landing in the 70s over a season is strong, and a single week at 20 per cent proves nothing either way.