You can connect through a locally running node, which reduces latency to some degree.

But warring state actors competing with each other on that offers me some protection.

Assuming they compete. If I were a state entity with a vested interest to compromise tor, I would cooperate with peers to that end, enemies or not. It is in every state's interest to have protocols in place for conditional cooperation with hostile states. At the agency or team level, these protocols can be quite effective.

After all, the field agents probably meet once or twice a year at some math/CS conference in France anyway.

And this is why governmental privacy is unethical... All should be open to peer review. For the people, and for the world.

I don't see how this would help. Such protocols may not even be written down, but rather implicitly passed from mentors to mentees in security agencies. I am all for government transparency, but no amount of transparency will reveal that a cluster in Utah is in direct link with a cluster in St. Petersburg is in direct link with a cluster in Kiyv to provide unmasking services to their administrators.

These administrators can then launder the information to their respective agencies by means of any number of play-pretend activities you can write up for the transparency committee. The agency doesn't even need to (officially) know.

Aren't there ways to filter out untrusted nodes?

(Edit: I say this, but in reality I also think it's pretty safe to assume most are government controlled)