lol, indeed. I had to set up a little ipython playground to actually test the various flags. There's like a weird double negative thing where you're like "Is failure=True going to fail or succeed if all the tasks fail" < mind explodes>
I merged my branch of apply3 to master, but it hasn't been tagged yet or pushed to conda. Check out what I did with step4 to handle cleanup and error messages. It didn't require using "failure" dependencies. Step3 is much more complicated, though.