Sex, race, some forms of disability etc. can all be relatively easily observed by another person, and therefore (with a brushing aside of issues of e.g. intersex and mixed-race people, how we define race, etc.) they're relatively uncomplicated to ascertain. If someone says "I'm male", I can look at them and say "no you're not, you've got a vagina". Similarly, for other types of disability or mental illness, if someone says "I have ADHD" I can check their behaviour against the behavioural diagnostic criteria for ADHD and make a (reasonably) objective decision about whether they're lying or not.
If a man says "I'm straight", I cannot look at them (or their behaviour) and say "no you're not, you have sex with men", because the definition of being a straight man is not 'never having sex with men' it's 'never being attracted to men' (which is an internal state of being, and therefore inaccessible to an observer). And whilst someone's behaviour (and self-reported internal state) can be a decent proxy for their "true" internal state, it's not the same thing.
A man who has sex with men may say "I'm straight", and be telling the truth; he's only having the sex because no one else is available, and he's imagining the other man is a woman the whole time, and is genuinely unattracted to his partner but 'making do' because he's horny. A man in an identical situation with identical behaviour may say "I'm straight" and be lying; he is attracted to his parter, but is in denial about it because internalised homophobia or whatever. How do I, an outside observer with no access to his internal state, say which is the case? Do I call this man gay, despite his self-reported sexual orientation, or do I call him straight, despite his observable same-sex sexual behaviours? (There's a reason sociologists often use the term "men who have sex with men"/"MSM", especially when discussing e.g. STD risks which are linked to sexual behaviour, not sexual orientation. No one in an academic environment wants to try making this call lmao.) Similarly, a woman who has never dated another woman and is in a happy opposite-sex marriage might say "I'm bisexual" and have it be true because she is attracted to both men and women; a woman in the same situation might say "I'm bisexual" and have it be false, because she isn't actually attracted to women and just wants to look trendy to her Bay Area friends; a woman in the same situation might say "I'm straight", and be lying, because she is attracted to women but in denial about it. Again, how the fuck do I know?
In all of these cases, I as an external observer likely have some evidence available to guide me in an educated guess as to whether the person is lying or telling the truth (e.g. past behaviour, things they've said, whether they're in a homophobic environment or not), but I cannot actually say for certain. That's where the definitional issues come in - especially when, as in the cases above, the available evidence is contradictory (i.e. people saying one thing and doing another). What do we call someone who performs same-sex sex acts, denies same-sex attraction, and who we suspect may be lying? What do we call someone who does not perform same-sex sex acts, claims same-sex attraction, and who we suspect might be lying? How do we arbitrate objectively on someone's subjective(-ish) internal experience?
This is, by the way, why science/academic environments often use weird tiptoe-y labels for stuff. Because they don't trust people to be honest (a lot of evidence shows people lie about or just straight-up don't understand their own internal states), but a lot of the ways we group people rely on internal states that researchers don't actually have direct access to. And so, the vaguely pomo language that's designed to either a) explicitly only describe someone's behaviour, or b) explicitly only describe someone's reported self-identity, to avoid the researcher having to make a call about whether the person is "lying" or not. This works okay in academia, because it's understood within this context of "hedging your bets" in order to accurately describe the group you're looking at; works poorly when it's exported into the regular world by psych undergrads with a twitter and delusions of intelligence, because it overcomplicates day-to-day shit to an unnecessary degree.