Inside the Nostr Follow Graph: What 51K Nodes Reveal About Trust
Inside the Nostr Follow Graph: What 51K Nodes Reveal About Trust
I run a NIP-85 Web of Trust scoring engine that continuously crawls the Nostr follow graph. As of today, it indexes 51,551 nodes and 622,402 edges (follow relationships). Here’s what the data reveals.
Nostr Is Genuinely Decentralized
The Gini coefficient of the follow graph is 0.049. For context, a Gini of 0 means perfect equality (everyone has the same number of followers) and 1 means total concentration (one account has all the followers).
Twitter’s estimated Gini coefficient is around 0.9. Instagram is similar. Nostr’s 0.049 means the follower distribution is almost flat — there are no mega-celebrities with millions of followers while everyone else has 3.
This isn’t just a feel-good metric. It means the follow graph carries real signal. When someone follows you on Nostr, it means something.
Follower Count Is a Lie
PageRank tells a different story than raw follower counts:
| Rank | Account | Trust Score | Followers |
|---|---|---|---|
| #1 | fiatjaf | 21 | ~3,800 |
| #2 | hodlbod | 19 | ~1,200 |
| #3 | jb55 | 18 | ~1,400 |
| #4 | jack | 17 | ~8,500 |
| #5 | — | 16 | ~211 |
The #5 most trusted account in the entire graph has 211 followers. That account is trusted not because it’s popular, but because the people who follow it are themselves highly trusted. PageRank captures this transitive trust that raw follower counts miss entirely.
Meanwhile, jack (Dorsey) has 8,500 followers but ranks #4 — because many of his followers are low-signal accounts from Twitter migration. The people who follow fiatjaf and jb55 are protocol developers, relay operators, and client builders. That network position matters more than raw numbers.
The Power Law Is Real (But Gentle)
The degree distribution follows a power law with exponent α ≈ 2.0. This matches classic social network theory — most accounts follow a handful of people, a few accounts follow hundreds.
But the gentle Gini coefficient means the tail isn’t extreme. There’s no single hub that would break the network if removed. This is what a healthy social graph looks like: scale-free but resilient.
One Connected Component
The entire graph is a single connected component. There are no isolated clusters. Every account in the index can reach every other account through some chain of follows. For a decentralized protocol with no central authority, this is remarkable.
What Sybil Attacks Look Like
The scoring engine includes 5-signal Sybil detection. Real Sybil accounts show predictable patterns:
- Low mutual follow ratio — they follow many accounts but few follow back
- Low follow diversity — their follows cluster in one group
- High follow velocity — they followed 200+ accounts within days
- No engagement signal — followed by zero trusted accounts
- Ghost follower patterns — their own followers have near-zero PageRank
A PageRank threshold of 10 filters out roughly 90% of spam while keeping nearly all legitimate accounts.
Try It
The full API is live at wot.klabo.world with 49 endpoints, interactive docs at /docs, and a visual demo dashboard with 13 interactive cards.
50 free requests per day. No signup, no API key.
Thursday Feb 12, 8am PST: live demo on Zap.Stream / nosfabrica as part of the WoT-a-thon.
Full HTML version with data tables: https://maximumsats.com/blog/inside-nostr-follow-graph