We all read people to an extent. Behavioural signs will show to a degree their emotional state, their status, their attention and interest in you/the conversation. Combine it with how they use language and you get a good idea of how they think and how self aware they are. You can see if people monologue at you or if they are interested in turn taking. You can get a feel for how quickly people can grasp information, how relaxed/restless they are, how internal they are, how nice they are, how insecure they are, how aggressive they are etc.

I suppose I also look also for how real a person is. For example in a work setting some people are much more prone to wear masks and fake emotions and some people don't do that. I do try to factor in how much games playing some people do/don't do.

From what I've seen, a few people are intuitively correct at reading others.

Unfortunately many people think they're intuitive regardless of how poor they actually are at reading others (high self-belief, but poor ability).

We all notice how it takes high skill to recognize the very highly skilled in areas we are talented in.

That was the less commonly talked about part of the Dunning Kruger Effect. While the Dunning Kruger paper has been somewhat dismissed now as due to statistical artifacts, the DK effect seems to resonate with real life so we want to believe it.

I imagine one gets a sense of how surprising the actions of others are. If you foresee what they will do, you know your model of them is accurate.

There’s the evergreen genre of business-oriented instruments that are mainly cheery self-assessments (“I notice when people are feeling sad: agree, or strongly agree?”)—but I’m not sure how valid those get.

As far as interesting efforts that reach for more objectivity, I feel like the personality psychology people circulated a bunch of “EQ” instruments like this in the ‘00s and ‘10s, rooted in Paul Ekman’s work on facial expressions:

https://greatergood.berkeley.edu/quizzes/ei_quiz/

Then again, they dealt with their replication crisis around the same time; I’m not sure how well these kinds of things held up.

Probably observing them longer and verifying if the predictions from earlier evaluation match their future behaviors.

Something like "she seems sad" and 5 minutes later "yup, she's crying now"