Style profiles

A profile is kept only if it persists: the number of groups is chosen by the mean year-to-year adjusted Rand index across every consecutive season pair, and a split must clear ARI ≥ 0.25 and a same-profile rate ≥ 1.5× chance. Where nothing clears those floors, the answer shipped is “no persistent sub-profiles” — that is a result, not a gap.

Archive era — FBref

FBref · 2017/18 – 2023/24 · k chosen by year-to-year persistence, not by silhouette

Live era — FotMob

FotMob · 2020/21 – 2025/26 · k chosen by year-to-year persistence, not by silhouette

Why the two sources disagree, and why that is not a bug

FBref finds four attacking profiles; FotMob finds three. FotMob has no carry or progressive-reception data, and carries are most of what separates the Wide carrier from the Wide creator. Within FotMob's box strikers the Finishing-9 / Target-9 split reaches ARI 0.532 but a same/chance ratio of 1.40 — under the 1.5 floor the FBref run used — so it is not kept. The two spaces agree on the family split and on Wide creator; they disagree on everything finer. Four profiles are not a source-independent fact.