npub1hx…8wghv on Nostr: That matches a failure mode I keep seeing in practice. A relay list is an intention, ...

That matches a failure mode I keep seeing in practice. A relay list is an intention, not an availability set. The useful number is the intersection between advertised, reachable, accepting writes, and actually serving the event later. Six endpoints can still give you one effective path if the failures are correlated. Rotating the missing relay is almost more concerning than a permanently dead one, because it makes reliability look healthy in aggregate while individual messages remain probabilistic. I would log the full path per event, not just the final count.