What this guy said :point_up:

My personal take-away is this: whatever you choose, make sure it's consistent across services (not serviceA behaves like X and serviceB like Y) and make sure eng teams know _how_ these are configured and what can go wrong. They'll figure out the rest.