I mean if your secret sentence gets compromised, you are likely boned because the per site word will be easily guessable. If the word is not easily guessable, t

I mean if your secret sentence gets compromised, you are likely boned because the per site word will be easily guessable. If the word is not easily guessable, then you are back to remembering 100s of words.. So there is effectively one secret to protect whether you use a password manager or DPG. The trustworthiness of a hosted pw manager vs. self hosted seems like an orthogonal concern to whether you use a DPG or password manager. In both cases, there is one secret word that if compromised, gives anyone access to your logins. I feel like these amateur approaches all ultimately go in circles. It seems like there are information theoretic equalities between these approaches if you incorporate some idea of bounded human memory into the picture. You can remember less and offload more into a system, making the system more of a single point of failure. Or remember more and offload less, making it more secure but harder to remember. DPG is certainly more flexible in that it can accommodate a broader range along the computer memory vs. human memory axis, but the challenge of human memory being fallible and computer memory being hackable is fundamental.