Many of them are badly translated articles. And all here is just hypotheses, which may be wrong. Some hypotheses are not even normal socionics like baseless Reinin's dichotomies, subtypes and other bs.
Also there is a problem of doubtful practice, as all have typing matches <50%, average <20%.
Types have specific expression in nonverbal behavior. You may notice the difference in my types examples.
"face structure" is just old heresy known as physiognomy. You may meet followers of this in socionics, unfortunally. As they also use behavioral theory - it keeps them in touch with reality.
Normal typing is IRL interview when you see the man. So you get there nonverbal info. As socionics accepts intuition as one of ways of knowledge, even equal to logics, - so "impressions" always were important in typing. One of things which gives us impressions is nonverbal behavior of the typed. You could meet in descriptions of types and dichotomies the "impressions" from people of some type. MBT should have this factor too as they use interview, not only the test, but probably they prefer to keep this way for typing as "hiden", as looking not too "scientific".
I did
comparative typing on socioforum with using nonverbal method - were typed bloggers with their random videos. Many typers have gotten match near 20%, what is much higher than 6% of random, and this was statistically correct (<5% mistake). This have proven nonverbal methods have objective basis, despite are mostly based on intuition.