In trunk based development, you do not use branches at all?

Of course you do - for PRs or just to self-review in some cases.

We do for pull requests going back to the trunk.

They commit incomplete changes directly to master after trying to make them unreachable. Personally I’d rather commit when ready, even if I’m going to be using a runtime experiment to ramp up slowly from zero. I don’t like being one mixup away from breaking prod.