Graphs for more than two: the whole view assembled

Second round for #463, built from Ross's answers to the first: "the half and half split for attributes and metrics & statistics is no longer viable as we are going to be showing a lot more data so I think tabs or chips for each one is better now so we only see one at a time", "I think all 3 radars are actually really useful and we could have all 3 on chips", "where the gap comes from is great", and "with the where the gap comes from we have 2 different views so can we incorporate both?"

Eight real defenders, latest capture. Every figure below — the axes, the category roll-up and the bars — is computed from their raw attributes, using RadarAxisSeedData's eight axes and ColumnDefinitionSeedData's three attribute categories.

the premise that expired

The halves were reasoned from a pair

#432 settled side-by-side halves deliberately, and the reasoning is on the issue: "the attribute half is a radar and eight rows and is done; the form half is a sample, seven or more rows and however many scatter cards." That was true of two players. A radar and eight rows does not describe eight small multiples and eight rows of eight dots — so the premise expired rather than the decision being wrong, the same way #628's eight-player cap expired with the dropdown.

One thing from #432 has to survive the change. Ross, then: "we need to keep this part at the top of the 2 player comparison no matter if you are graphical or data or you would have to keep switching between them to change your comparisons." The bars are how you pick who is being asked about, so they stay pinned at the top of whichever tab you are in — the tab swaps what is under them, never them.

the proposal

The view, at the width it now gets

One kind at a time, full width. On Ross's screen that is about 1625 rather than the ~800 a half was getting.

Click a player to ask about him — the bars stay put whichever tab is open (#432)
Radar A radar each Against the field Overlay Scale 0–20 Fitted

Where the gap comes from

By axis

Against The field's median The best of them
The anchor Everybody else Ahead of the reference Behind it Hover any dot for the name
Ross's fourth question

Gap and Levels in one row

Ross: "with the where the gap comes from we have 2 different views so can we incorporate both?"

They answer different halves of one question and neither is complete alone. Levels says where the men actually are — and a gap of 1.0 at 16 against 15 is a different fact from 1.0 at 9 against 8. Gap says how far apart and who leads, which the dots imply but never quantify or name.

So the row above carries both: the track is the levels view with every player on it, the coloured segment is the gap, and the figure on the right names it. The toggle disappears — there is nothing left for it to switch between.

And it degrades correctly at two players With exactly two men on screen the reference is simply the other one, the track carries two dots, and the row reads as today's Gap view with the absolute positions added — which is today's Levels view. So the pair case does not become a special case; it is the same row with a field of one.
Settled — the reference is the median #463's question 6. A gap needs something to be a gap from: median is where the group sits, best is the bar to beat, and they tell different stories — Maguire is above the median on Aerial and still behind de Ligt. Ross: "median". Best stays on a chip, because "can he reach the best of them" is a real question, just not the default one.
Settled — the scale is 0–20 Fitted gives about 1.8× more separation — every figure in this squad sits between 7.8 and 17, so a 0–20 radar leaves its inner 39% permanently empty and squeezes Physical's whole spread into 7% of the radius. But fitted is relative: the scale comes from who is on screen, so adding or removing a player redraws everyone, and a shape that fills the radar only means "best of this bunch".

That would break a rule the rest of the app keeps — the Analyse screen and the comparison filter both insist that narrowing what is listed never moves what a man is worth. Ross: "0-20". Fitted stays available, and earns its keep on the small multiples where you are reading shape against shape and the cards already name the set.
Settled — the three radars, and what each one is A radar each is the anchor against each rival — seven cards, not eight. Ross: "with de Ligt selected you would [have] a comparison of him against all the others." That is the better shape because it is today's radar repeated: the anchor blue and the other man white, the two colours #409 already spent, with nothing new invented.

Against the field is the one-radar reading, and overlay is the deliberate two-or-three mode. Only overlay costs new colour, so it is the one that caps itself — at four, and it says so.
Measured — the other three cannot be chosen silently Ross: "the overlay is capped at 4, how do we choose which 4?" Measured across the eight anchors, the candidate rules agree on only 9 of 24 picks. For de Ligt: nearest in the order gives Maguire, Yoro and Heaven; most like him gives Min-Jae, Maguire and Martínez; least like him gives Mazraoui, Shaw and Heaven. Three different answers, so any silent default would be making a choice the reader cannot see. It is four chips, defaulting to nearest in the order — the reasoning the current stopgap already uses, that this is the contest a depth order decides.

One caveat on that default. The order here is a stand-in — the mean of the eight axes, not the role-weighted score the app ranks by — and on this squad it puts all eight within 0.99 of a point. The disagreement finding holds regardless, because "like him" comes straight off the axes, but the quality of the order-based rule cannot be judged from this mockup.