I use an algorithm. In short: 1. Memorize some base password 2. Memorize a way in which you mutate that password based on the name/type/other of the service logging in to. Eg. Hunter2 becomes eHunterG8 Because my example algorithm cares about Google's first letter, length, and service type: email. It allows every password to be different but you only memorize two things. It is meant to be a "good enough" solution tha…
Does your threat model include anything NOT bulk, ie where a human would see a password and guess the pattern?
Have you tried any of the password managers that have integrations, to automate password changes in case of failure?
Have you ever heard of the security concept of minimizing your attack surface?