I think this is a bit of a red herring. ULIDs are not meant as a replacement for a lamport timestamp or something. The accuracy to the millisecond would only be

I think this is a bit of a red herring. ULIDs are not meant as a replacement for a lamport timestamp or something. The accuracy to the millisecond would only be correctly ordered if the clocks were perfectly synchronized anyway. I interpret the purpose of the timestamp based id as allowing denser clustering in b-trees vs random key data, and that’s about it. The timestamp can only be used as a relative ordering when processing multiple events from the same client, and even then not completely (client clock could change).