Turns out consistency and latency metrics on trustedrelays.xyz were penalizing a large number of relays unfairly (particularly specialized relays for various reasons) so pushing an update to move from standard deviations to interquartile range to smooth things out. Should see quite a few relays get an increase in overall scores as a result.
Lesson: math is fun 🤣🤣
Huge shout out to anyone running any sort of Nostr infrastructure. It's a thankless job and the network wouldn't run without it. Some observations from the last few days, we need more NIP-66 monitors, particularly in the Americas.
If you run a specialized relay, probably worth going over your NIP-11 data to make sure it reflects what your relay can and cannot do for transparency to your users. Seeing a lot of relays with just the very basic information which is fine, but if relay trust assertions ever gathers steam, the basics just won't be enough.
With that said, just updated nip46-relay (forked from Bunklay) to be more NIP-11 spec compliant by adding supported kinds, and relay limitations. Changes are live at relay.nip46.com as well.
Pushed some scoring improvements to Trusted Relays today:
- Recent reliability matters more than history (3-day half-life)
- Frequent short outages now penalized, not just long ones
- Offline relays decay gradually instead of flat cap
- Stacked access barriers (auth + payment) use diminishing returns
- Stale NIP-66 monitors excluded from calculations
Also revamped the Insights section with operator reputation context, comparative rankings, and trend analysis. Changes are live now and scores will start reflecting over the next day or so.