Why the opponent matters more than the player

“How has this player gone against this club?” is the first thing most people check and the thinnest number on the page.

Set the two side by side. A head-to-head is two or three games, sometimes one, occasionally none. What the other club gives up is counted over every game it has played this season — 23 of them, at the time of writing. That is the whole of the argument in the title, and it is worth being precise about what it does not say: nothing on this site measures which input predicts a result best, and no page here claims the conceded table does. What it claims is that one of these two numbers has a sample and the other does not.

The head-to-head is a split like any other

On a player's props card the head-to-head sits in the Splits row with Home, Away, Wet and the rest, and it obeys the same floor they do: below three games in the window on screen the chip is not drawn at all. The projection treats it the same way on purpose. A career head-to-head appears in the working as evidence, with its average and its game count printed — and it is never added to the number. Two-game samples fail a four-game floor more often than they pass it, so the projection shows the head-to-head to the reader and adds nothing for it.

What the feed records, and what we build from it

Every player-game row we hold carries the opponent stamped on it. That part is recorded. “Conceded” is not: no feed publishes a table of what each club gives up, so we work it out from the game rows. For a given market — disposals, say — we sum every opposition player's disposals in each of a club's games this season, divide by the number of those games, and sort the eighteen clubs from most conceded to least. A club's place in that order is its rank, and rank 1 means concedes the most. A club with no game carrying the key is dropped from the table rather than ranked last, because a club we hold nothing for is unrankable, not the stingiest defence in the league.

That is one derived number with one denominator, and both are printed wherever it appears. It is worth saying plainly what the word derivedis doing there. Disposals are recorded: a feed publishes them, per player, per game, and we store what it publishes. The conceded table is arithmetic we perform on those rows and nobody else's number. Anywhere the two appear together on this site, the derived one is labelled.

Where it shows up, and what each part costs

Four surfaces read the same table. On a player page, the Matchup Analyser island names the club, its per-game figure and the league average, with a chip giving its rank out of the eighteen; that island is on the Yearly plan. On the props card, the Opp Rankings row turns the table into a filter — narrow it to the third of the league that concedes the most and the game log redraws to the games against those clubs, which is a real sample about a real property of tonight's opponent rather than a two-game head-to-head. That is free, and so is the panel behind it, which draws all eighteen clubs on one axis with the league average marked. So is the projection's opponent row. The compare board's per-column version is Yearly again.

What the numbers looked like on the day this was written

Across the 2026 AFL season, with 23 games behind each club, the most-conceding defence gives up 395.6 disposals a game and the stingiest 344.3. The mean of the eighteen is 368.4, which is the figure every leak on the site is measured against.

Work one through. Take a player averaging 24.0 disposals a game facing the club at the top of that table. The club concedes 395.6 against a league average of 368.4, so 27.2 more disposals are available in the game than in an average one. The player takes 24.0 of a league-average 368.4, which is 6.5%of a club's disposals. 27.2 multiplied by 6.5% is 1.8. That is the whole adjustment: plus 1.8 disposals, in disposals, arrived at from four numbers all printed on the page. The ceiling on any single adjustment is a quarter of the baseline — 6.0 here — so it is nowhere near binding.

Why a small spread is still worth reading

344.3 to 395.6 is a spread of about 15% across the league, which is not dramatic, and a reader who expected the choice of opponent to swing a projection by a third should recalibrate on that number rather than on intuition. What makes it worth reading is the other half: the projection caps every adjustment precisely because a thin player-side sample can manufacture a figure, and the conceded row carries no sample floor at all, because it is a whole-season club total rather than a handful of games. It is the one number in the projection whose sample is never thin.

How to read it

Read the rank with its receipt — the place out of eighteen, and the games it was counted over. Read the leak as what it is: a difference in the stat's own units, from the league average, scaled to this player's share of it. Read the head-to-head beside them as evidence, and notice that the page has already declined to add it.