Weirdly I have no problem at all with the Mind in the Eyes test; dunno what exactly I scored, now, but I remember that it was better than the average non-autistic.
But the thing is, IIRC they're pretty much pictures cut from magazines, I guess actors or models or similar, trained in using their faces to convey particular mental states on demand, using both universal human expressions and culturally-specific and art-form-specific conventions. And the person is often chosen for a job because their face and eyes happen to almost inherently convey particular emotional states (e.g. someone with "sad eyes" which just look like that anyway, because of the eyelid shape). They've been instructed to perform a certain scenario/feeling/whatever, and they're consciously, deliberately and non-spontaneously configuring their faces to communicate a mental state, in a way that will come through even through a still photo.
I think, if I'm remembering right, they then got some non-autistics to interpret the meaning of the eye cutouts they selected, to produce a "correct" answer, and then added some wrong answers for the test? Basically, "reading" these eyes isn't much like interpreting faces/eyes in real life (those ones move , and are in 3D, and are attached to a whole person, and aren't portraying a simple character for a media product). You don't get the "correct" answer by being able to tell what the person is thinking, but by being able to read a stylised performance in the same way the majority do. A matter of cracking the code for which facial arrangements are conventionally used and perceived, in our society's media, for communicating the impression of which mental states — regardless of any actual inner mental state. It seems to test how closely autistics can approximate the interpretations of non-autistics, of a particular type of media. You might as well do it with Noh theatre, or any other human creative output with particular conventions.
I know for some autistics this is still incredibly difficult, or even impossible, no easier than reading real emotions and mental states on the faces and bodies of real human beings in front of them. I have family members who are incapable of reading any but the most obvious facial expressions, and cannot detect fakery at all. But for me, the media conventions that Mind in the Eyes tests knowledge of and ability to interpret are things I find much easier than actual real people's actual real nonverbal communication.
One of my ✨things✨ has always been language and communication. Like a lot of people, when younger I tried to analyse the fuck out of how human beings, more specifically human beings in my society, intentionally and unintentionally encode thoughts and feelings in the position, tension and movement of facial muscles, the movement of skin and eyes, positions of various body parts, and other nonverbal means. I paid attention to media portrayals, took note of other people's use of the codes, looked at myself in the mirror while practising conveying messages with my face, read as much as I could about the subject, learnt to convincingly fake a Duchenne smile, and so on.
I'm still not great at interpreting the involuntary stuff, or seeing the subtle communication behind the surface layer one, but the Mind in the Eyes test plays right into one of my most lasting fascinations 🤣