Makes me think I need a better strategy on the username side to not leak that info.
Even if the dictionary and patterns are known by an attacker, this still has very good entropy on ~16 characters long sentences. For usernames you can easily just go down to the minimum (around 6-10).
I find this superior to random values in a lot of places as it's easily readable and it's also easily typeable, when copy-paste doesn't work.
[1] https://bitbucket.org/ligos/readablepassphrasegenerator/wiki...
- if they start sending you spam you can severe their capacity to contact you by deleting the alias
- if they give your contact to a third party, you know from the alias who leaked your email address
- if you see an email on a data breach like this one, you know immediately which website got hacked
- it makes it really hard to correlate your identity across two websites
How does that follow? It still adds value to use aliases that identify the site where they're used, because then you don't need to do a hashtable lookup to see where your mail is coming from.
What I and many others do is reuse not a password, but reuse a password formula (in my case with slight but easily memorable variations depending on importance, or if forced by stupid password rules). If you saw a dozen of my email/password combinations in the clear you would be able to reverse engineer the rules and then guess combinations for arbitrary sites almost as good as me. It's a calculated risk, just like trusting a password manager is a calculated risk. Right now I consider password managers the better trade-off, but still only slightly better, by a margin small enough to make it not worthwhile to invest in a habit change. If the "formula" I happen to use was just a little harder to reverse engineer than it sadly is I would consider it strictly safer than password managers.