I hadn't heard about CUIDs, but based on the Github repo [1], I see they fall into two of the categories I presented:
- A large range. It looks like CUIDs support a short version, but the README calls out the short version should only be used for disambiguation, and are "absolutely not recommended for database unique IDs".
- Encode the partition in the ID. In this case, the partition is either based on the browser or the process ID of the server.
