A seeming illogical design\test strategy?

The main reason we don't set experiments per user is because signing into the browser isn't required. Whatever system we use needs to work regardless of a user's identity, so in order for experiment results to be statistically valid, we can't change the way we set experiment populations for any reason, including whether or not the user has an identity that is known across other devices.

For what it's worth, the best way around this is to use flags for any feature that you want to explicitly turn on, although that does require that we actually create said flag.