Most fantasy statistics describe what already happened. Only some of them tell you anything about what happens next — and those are the only ones worth building a team on.
The distinction is simple to state and easy to forget: points scored is a result, not a predictor. A player on a huge score last week has told you almost nothing about this week. What he does repeatedly, on the other hand, tells you a great deal.
The numbers that hold up
Minutes. Before anything else. A brilliant player at sixty minutes a week is worse than a decent one at ninety, and no other statistic survives contact with a benching. When I look at a player I haven't followed closely, minutes are the first column, not the last.
Expected goals and expected assists. These measure how often a player gets into positions that produce goals, rather than whether the ball happened to go in. Actual goals swing wildly around the underlying number in both directions; the underlying number is far steadier, which is exactly what makes it useful for forecasting. When a player's goals are well below his expected goals, the usual resolution is the goals catching up — not the chances drying up.
Shots, and where from. Volume in the box is one of the most repeatable attacking signals there is. A player taking regular shots from close range will score; a player taking speculative efforts from distance mostly won't, however impressive one of them looked.
Touches in the opposition box. My favourite single number for attacking defenders and wide players. It answers "is this player *involved* where points are made" without waiting for a goal to prove it.
Set-piece duty. Not a statistic exactly, but it belongs on this list, because it's the most repeatable source of chances a mid-priced player can have.
Defensive actions, now that they score directly. Tackles, interceptions and blocks are unusually consistent from week to week — they're a product of role and position, not form — which makes a defender who reliably clears the threshold one of the more predictable assets in the game.
The numbers that mislead
Total points and form. Backwards-looking, heavily driven by one or two big hauls, and the thing most likely to make you buy a player at his peak price after the returns have already happened.
Goals, read alone. Over a short window, goals are mostly a story about finishing luck. Read them next to expected goals or don't read them at all.
Bonus points, read as a skill. Bonus follows from the underlying actions — shots on target, key passes, defensive work — so it's better predicted by those than by past bonus.
Season totals with no minutes attached. A player with twenty starts and a player with thirty-five are not comparable on a raw total, and this is the single most common way managers overrate a nailed-on starter and underrate a better player who missed two months.
Why per-90 changes the picture
Dividing by minutes played is the simplest correction in fantasy analysis and it reorders lists constantly. Per 90 minutes, the substitute who's been quietly excellent stops being invisible, and the ever-present who's been merely present stops looking prolific.
Two cautions. Per-90 numbers on small samples are unstable — a player with 300 minutes can top any per-90 chart by accident. And per-90 answers "how good is he when he plays", which is only half the question; you still have to multiply it by how much he actually plays. The player you want is the one who's strong per 90 and certain to be on the pitch.
How I'd actually read a player
In this order:
1. Minutes and role. Is he starting, and in the position where points come from?
2. Underlying attacking output per 90. Expected goals, expected assists, shots, box touches — is he involved?
3. Set pieces. Does he have a repeatable chance source?
4. Fixtures ahead. Over a run, not a week.
5. Price and ownership. What does he cost, and who already has him?
Points scored comes last, as a sanity check — not as evidence.
The honest limits
Underlying numbers are a better guide than results, but they're still a description of the past. They don't know about a tactical change, a new signing arriving in his position, or a manager who has quietly lost faith. A small sample can say anything, and early in a season every sample is small — which is precisely when the temptation to read them hardest is strongest.
The live version is the Premier League player stats board: season and per-90 views, filters by position, price and team, and the underlying numbers next to the points they actually produced. For how those inputs become a single comparable number, see FPL expected points explained.