@darius I am so often a champion of “you can’t learn if you don’t ship” and “build safety, then experiment constantly” approaches, which really start to fall apart when your product is basically a giant psychometric challenge trial (with no IRB, of course) on the global population. But I still feel like “stop moving because moving might break something” is just as bad when the stakes are high as when they’re not. Constantly wondering how to resolve this tension.