Your criticisms seem mostly on-target, but I don't think you make a very good case for why it matters.

That a two-cluster model of sex explains a heck of a lot of variance is inarguable. That it's biology, so everything is messy and you have all sorts of diversity, is also inarguable (and Stock doesn't really even try to argue--otherwise the cluster account wouldn't be necessary). What of it? Sexual differentiation is heavily regulated with mutually inhibitory feedback loops, and thus is one of the cleaner instances of bistable states. But, biology! You can find (rare) instances of all sorts of genetic variation that fails to preserve the regulation.

In terms of typical evaluation of efficient use of language or of theory for capturing an intrinsic feature of distributions of phenomena observed in the real world it's absolutely not the case that an octonary, quintary, or unary conception is as good as a binary one. Admittedly, as a practical matter, we needn't care about the difference between one bit and three because we generally use piles of bits to encode our one-bit distinction (even with, say, -o vs -a in Spanish you still use 5ish bits of entropy that you need for an extra character to encode a 1-bit model...and then you do it on tables and air (and differently, depending on how fast it's moving!) and stuff, for which the model provides roughly zero explanatory value). But that it's workable to have not-two categories is scant praise. Two categories for sex is a good model, and HPC--from your description of it--sounds like a pretty workable way to justify the two categories if you, for some odd reason (e.g. because you're a philosopher and it is the philosophical tradition to do so), want to go to a whole lot of trouble to note that something that is abundantly supported by evidence is, in fact, there.

So, okay, we have Stock saying "binary model with noise", and your best motivation against it seems to be, "because it's a standard pretext for harmful anti-trans campaigning".

That doesn't make the model wrong.

I don't see the issue. If I'm a noisy data point, it certainly doesn't help me for Stock to observe that I'm part of the noise in her model--I'm still a person, and advanced societies have lots of surplus resources to devote to making things work out well for people whose concerns occur with low frequency. Not very many people are actually gluten intolerant, but we've switched over to making things waaaaay easier for them. (Admittedly, partly out of a fad to think gluten is more problematic than it is.) Peanut allergies, ramps for accessibility, color blindness, and so on. We handle a lot of "noise".

It seems here that you're trying to fight Stock on principle rather than just shrugging and accepting the point about reality--"fine, if you want to call sex binary with noise, that's cool...now, let's talk about the noise (if we classify gender incongruence as part of that noise)".

Indeed, fighting the wrong battle here is one of the places where trans advocates I think do the greatest amount of damage to their cause and their credibility. You don't deny that there are three phases of matter (at standard temperature and pressure) just because porridge and butter. But when you want to talk about the physical properties of porridge, you need to take seriously what it is.