Conversation

Replying to
I’d also put a plug in for needing intentional, random assignments of differing treatments for most large scale data. Hard to figure out what to do just by looking at history without tests.