This is a very good article overall but here and in a variety of other places, there is an over-reliance on categorical rather than quantitative thinking, which can make a reasonable position seem unreasonable.

For instance, "If you can still die in a car crash while wearing a seatbelt, why have seatbelts?"

Well, obviously, because it's not a qualitative issue. It's a quantitative one.

Are violent women who are in prison as violent as violent men (who would pretend to be trans) who are in prison?

This also comes up in your reasoning about incercernation of people for nonviolent crimes, cultural misogyny, cohabitation of people convicted of nonviolent crimes, frequency of trauma-induced fear, rate of entry of the opposite sex into same-sex spaces, etc..

Nonetheless, most of the time the issue is so clear anyway that it doesn't make very much difference to the conclusion.

I am puzzled, though, about why there's any difficulty in solving the shelter issue.

You have women's shelters for women: cis women and trans women who have fully transitioned (probably--if it turns out that trans women need different support, then they could have different facilities). That will handle the majority of the need with minimal exacerbation of trauma. Then you have other shelters for other groups. The different shelters should be networked so people can show up at the door of anywhere, and be taken (for free) to a place that is designed to be able to keep them safe and start getting them the help they need. The fundamental principle, though, is that people who have experienced trauma shouldn't have to expose themselves to trauma-triggers or be accessed by the person/people who are perpetrating the trauma, so you subdivide things until that condition holds an acceptably large fraction of the time.