The documentation of this common complaint seems a bit scarce, so I'll expand a bit on it. There are many signals feeding the algorithm, but this only leads to confusing results with too many results that are unlikely to be relevant (The goal of this is discovery, but the prioritisation and results are not following the expectations of the users).

I'm using the language picker here, instead of the article language/short picker, because it is more reproducible:

A practical example: For an English language browser, for users based in the Netherlands, the suggestion list includes things like: "Bahasa Indonesia, British English, Deutsch, English, français, Frysk, Limburgs, Nederlands, Plattdüütsch, Tarifit, Türkçe, Zeêuws"
All languages commonly spoken in the Netherlands, the (former) territories of the Netherlands, or derived from Dutch.

Yet most of them are not relevant to most people and the most likely one: "Nederlands", is even below the fold:

This should probably be ordered as: browser language, official local languages in order of size (nl, fr, de, fy) (assuming Netherlands and Belgium collate), secondary local languages (dialects), the rest (Turkish, Indonesian) instead of being alphabetically ordered. (Kinda surprising that Turkish and Bahasa Indonesian got added, yet no one added South African. Additionally, there are probably more speakers of Ukrainian in the Netherlands, then of zeeuws, so this region based suggestions is always a very complex and variable method)

Now Compact Language Links adds to this, that in order for the language to be shown, it has to actually have an article in that language. This creates the problem that the list varies for every single page and every single time you will have to guess if the language is not shown because the article doesn't exist, OR because it didn't include your language as a suggestion. This is a UX problem of 'stability' and 'predictability' leading to bad 'scanability'.

In the full list, you have a lot of entries, but you know exactly what those entries mean. In the compact list, the presence of a language means TWO things (it exists AND it *might* be interesting to you). This can help discovery of a language, but it is terrible at answering the reverse: "does a specific language I'm interested in have an article on this"

Found this via the proposal. Please also allow specifying which languages are shown at the top so it doesn't change every now and then. There could also be a field above Suggested languages for these configured languages which could e.g. be all the languages one understands to some extent (or all languages of WPs that one frequently uses for machine translated things where eg the WP has relatively high quality/coverage and MT is good for the lang).