If you're going this route, I highly recommend looking into and using OKLAB instead.
The problem with HSL/HSV is that it's not perceptually uniform - if you only move HUE to change color, you will get different perceived brightnesses. This is important especially when procedurally generating color palettes, but also makes it harder to pick a color.
OKLAB solves that issue, and is designed to be uniform. Here is a great article about it, which is funnily enough IIRC a blog post that invented the color spectrum, that got noticed and eventually turned into a new industry standard.
Here is a picture that sums up pretty obviously what is the difference. This is a gradient that moves just the hue.
The algorithm is probably made to maximize the time you spend on the platform, and is really good at it. (I mean, just look how good are ML algorithms on text -> picture, and add to it that the algorithm that does your info -> engagement has decades of data and training on billions of people).
My theory is that it has misaligned, because it turned out that radicalizing people into right-wing bullshit will glue them to the social network very effectively, so it just started to do that. It makes sense - once you start spewing right-wing bullshit, it will probably isolate you from your IRL friends, you will have an echo chamber on the social network, and it is made to sound like some kind of deep truth no one else knows.
You getting left-wing content might be simply because it would not be efficient to try to convert you, so the algorithm is trying something else that's more effective on the (minority?) of people like you.