Looks good! I like that it’s deterministic by piggybacking off of BIP32 derivation. Seems to cover my use case. 👍
Out of scope for the NIP, but just thinking out loud: an aggregator of invalidation events could publish a bloom filter of invalidated npubs. There could be false positives, so clients would want to query for the specific invalidation message.
Yep. I think statistical filters will be important to ease per-client effort. However it means you’ll need to double check for positive results.
Event clients sharing a filter for what events they already have and a relay filtering them out or similar.
I wonder if a compound filters have event been used, where it’s still never 100%, but you have two offset filters and it would need to be true in both.
