I am running outbox benchmarks to increase confidence in 1) baseline outbox algos, and 2) algos with thompson learning.
Current understanding:
1 year outbox coverage
Filter decomposition (rust-nostr)
25% coverage without thompson learning. 37% with thompson learning (+48% coverage)
NDK
20% without. 31% with thompson learning (+55% coverage)
Coracle/Welshman
30% without. 39% with thompson learning (+30% coverage)
Should have more on Monday.
https://github.com/nostrability/outbox/pull/35
cc @hodlbod @PABLOF7z @hzrd149 @Max @Awiteb @Mike Dilger ☑️ @Yuki Kishimoto @KernelKind @ @Alex Gleason
