RFC 9562 says UUIDv7 implementations MAY alter the timestamp and makes no guarantee about how close it is to the actual time. Postgres docs also warn the extracted timestamp may not equal creation time. A post on alexsci.com shows Python's uuid.uuid7() running 3,695 s ahead of the clock after a one hour rollback, and growing under load. A CPython developer says to assume the system is flawed.
Built on Shards of SiliconUUIDv7 doesn't have the properties you think it doesWhen timestamps aren’t time-ordered and other oddities