> If you do something obvious like use amazon@example.com at amazon, you should expect them to add a rule that blah_blah_anything@example.com gets cleaned up to "@example.com"

It seems like a very error prone thing to do automatically, while reaping little in return for the effort. How much do advertisers stand to gain by adding special case heuristics for "people who might be using an unshared email domain with a per-site local-part?"

If I give my email as me+amazon@gmail.com, normalizing it to me@gmail.com is just applying a bit of common knowledge about how an email service with billion(s?) of users handles the + sign in the local-part of an address. It's a carve out, but it's totally deterministic (for Gmail at least) and applies to a lot of users.

If I give my email as amazon@2d0d4809.com, normalizing it to @2d0d4809.com involves guesswork about that specific domain. I don't think normalizing away the + sign in the first case implies we should expect the latter to get normalized into a single identity as well.