Graphs for more than two players: the attributes radar

For #463. Ross: "the radar is only 2 players and the where the gap comes from is also only 2 players... an idea could be to choose the players we want in the comparison, ideally more than 2 players on the radar would be useful as long as you could tell who is who. Another option is to have a radar for each player in the comparison as well as the where the gap comes from."

Both ideas are drawn below, with a third the measurement suggested. Eight real defenders, latest capture, put through the eight outfield axes exactly as RadarAxisSeedData declares them — every composite here is the mean of that axis's attributes, skipping any the player has not got.

measured first

Why this is harder than adding polygons

The obvious move is to draw more shapes on the one radar. The figures say why that stops working fast — and it is not simply "too many lines".

The eight defenders, on the eight axes

The spread on each axis

The polygons are not spread out — they are nearly on top of each other A 0–20 radar spends most of its radius on values nobody has. On Physical the whole squad fits in 1.4 points, which is 7% of the radius: eight polygons within a couple of pixels of one another. That is the problem to solve, and adding colours does not solve it.

This app has already met it and settled it once. SmallMultiplesConverters, for the Player development graph, is "deliberately scaled to this attribute's OWN min/max, not FM's shared 0-20 range — the whole point of small multiples over the one combined chart it replaces is that a Stamina swing between 1-3 and an Agility swing between 12-13 both read as a visible shape instead of the low-range attribute flatlining." The same reasoning applies here, and the same shape answers it.
Ross's first idea

A — More than two on one radar

Pick who goes on. Add players below and watch where it stops working — the point of drawing it is that the breaking point is something you can see rather than take on trust.

Scale

What it costs

It needs colours, and #409 spent them. Blue and white say who, and the data view settled the rest as anchor blue with everyone else neutral. A ramp per player re-opens that, and the legend becomes the only way to read the picture — you are looking things up rather than seeing them.

It works at three. Four is arguable, five is a hairball, and at eight the shapes cross so often that no colour tells them apart. Switch the scale to fitted and it gets better, which is the hint the next shape takes.

Worth having as a deliberate overlay these two or three mode. Not as the answer to a field of seventeen.

Ross's second idea

B — A radar each, against the field

One small radar per player. Each draws that man solid, over a shaded band showing where the rest of the field sits on every spoke, with the field's median as a faint line. You can tell who is who because each card is labelled — the thing the overlay has to spend colour on.

Scale Backdrop
This is the one I would build It answers both halves of what Ross asked for at once — a radar for each player and you can tell who is who — and it is the only shape here that does not get worse as the field grows. Seventeen centre backs is seventeen cards; seventeen polygons is nothing at all.

It also spends no new colour, so #409 stands: every card is the same blue, because the card's heading says who it is. And it lets you do the thing neither of the others can — scan for shape. Mazraoui is visibly a different animal from Maguire here without reading a single number.
the measurement's own suggestion

C — One radar: the anchor against the field

The shape the data view already takes. One man solid, the field behind him as a band — its range on each spoke — with the median drawn through it. Scales to any number of players because the field is summarised rather than drawn.

The anchor Where the other seven sit The field's median

What it answers, and what it does not

Answers: is this man ahead of the group, and on which spokes. At a glance, on any size of field, with no legend.

Does not answer: who is who. The band is everybody and nobody — if two men are pulling the top edge in different places you cannot see it. This is the shape for how does he compare to the rest rather than compare these men.

#463's question 6 lives here: the band needs a definition. Range says "the spread you are choosing within"; median says "the middle"; best says "the bar to beat". Drawn as range + median because that is the pair that does not need a caption — but this is a real decision, not a default.

the other half of the question

Where the gap comes from, with more than two

The gap view decomposes one pair. Against a field it has to become something else. A row per axis with a dot per player is the shape that survives — #463 already guessed as much, calling the level rows "the graph that scales most naturally", and drawing it bears that out.

The anchor Everybody else Hover a dot for the name · the bar is the field's range on that axis
It keeps the gap without needing a pair Each row is one axis on its own track, so the bunching that ruins the overlay is exactly what this shape makes legible — a narrow band is a short bar, and the dots inside it are still separable. It also reads at seventeen players, where the only cost is dot crowding on the tightest axes.

The category roll-up is the piece that genuinely does not survive: it explains a gap, and a field has no single gap to explain. Either it follows the anchor against the field's median, or it stays pairwise and you pick the pair. That one wants deciding rather than drawing.
what I would do

The proposal

B as the default, with A as a mode. Small multiples answer the question Ross actually asked and are the only shape that does not decay with the size of the field. An overlay control on top of them, capped at three or four, covers the head-to-head case the pair view does today — and because you chose those three by hand, a legend naming them is honest rather than a crutch.

C stays available as the one-radar reading, and is what a single card in B is already showing.

The levels strip replaces the pairwise axis rows, and the category roll-up is the open question.

And one thing that is not about the radar at all Every figure in this squad sits between 7.8 and 17.0 of 20. Whatever shape wins, a 0–20 axis spends 39% of its radius on values nobody has and squeezes the differences that decide the comparison into the outer third. The fitted toggle above is on every shape here so the difference can be seen rather than argued about. This is the same call the Player development graph already made, and it is worth making deliberately rather than inheriting FM's scale because it is FM's.