I usually work off of self-report data. I listen to people, read what they say, I look at how they phrase their communication and what semantic patterns they tend to employ.
right, you do focus on observable properties. this is the broader concept of what i get at with "behavior". behavior is not of special importance in this regard.

I'll also examine their values and beliefs, desires and motivations, what they're reactive to, their outlooks on the world, etc.
this is more difficult, because to reach these things one needs to engage in an act of interpretation. usually this interpretative step can not be fully revealed to an audience. this is why the practice is subjective and prone to creating controversy.

Ultimately, what's vital from all of this is getting an idea of how the person perceives themselves to be and through what sort of psychological orientation they tend to experience life and reality.
this "getting an idea" is what i'm referring to. there is always something left unexposed about how these ideas are formed. when another person sees eye to eye with you on the judgment, this is not a problem. when the intuition is not shared, you need a more "public" way of explaining the argument, and it is here that usually observable characteristics are cited.

When and how does information metabolism inculcate stereotyping?
Se is willpower, Ti is logic, alpha is fun, etc, etc, etc...

And certainly, even if it did, it's been nowhere near as egregious of an abuse as behavioral stereotypes always have been.
the practice is equally commonplace if not more so.

What interpretations are you referring to? Point at something.
i don't have time. the argument is really basic. it's amazing you could even suggest that behavior is more variant to interpretation than socionics functions and IM processes are.

Usually when referring to a person's 'behavior' in a psychological context, this is taken to mean something about that person's volitional actions or conduct. Whereas things like autonomic physiological processes wouldn't fall under the same umbrella.
behavior is not the primary issue. i'm mainly getting at anything that can be observed and pointed out without raising controversy over it's identity. such properties can then be used as undenyable bases for consensus formation. IM processes occur "behind the scenes", on an unobservable plane, and are as such never instances of such properties.