Core message of Pavel's talk: if a system can reach a given state through two different paths and you can't tell which path was taken, you've lost data. You can consciously drop data later, but unintentionally losing it is dangerous because more often than not you didn't know you should have stored it. Example: e-commerce teams obsessively analyze abandoned carts but rarely analyze items added and then removed before finalization — losing valuable customer information in the process.