> The notice may be from a commercial entity but it's court-ordered. It's not spam.

How is the filtering algorithm expected to know that? Especially if numerous users do mark such messages as spam (or give the more passive signal of completely ignoring it despite paying attention to other messages), or other identification rules say that the messages look like other things that have been thusly marked over time?

> those automated algorithms based on feedback need to not cross user accounts

One of the touted advantages of collective mail systems like gmail is that such filtering can apply globally instead of us all having to individually train everything to our liking. There are conflicting priorities, and unfortunately your preferred priority just isn't winning here.

[Caveat: I don't use Google's mail services for anything other than occasional testing, like sending messages to/from my own mail server after reconfiguration or other admin work]