How jumblewisp scores trust

npub1vlprg9j8u5l92az0zd6yd8ks7tl560v8ssepdkn07nwekdl9rs4saccfwp
hex
3054a6c465a19ea3c1e76c0b9e240275975e8e863ee2e608650bd45d13038b1anevent
nevent1qqsrq49xc3j6r84rc8nkczu7ysp8t96736rrachxppjsh4zazvpckxsprpmhxue69uhhyetvv9ujuem4d36kwatvw5hx6mm9qgsx0s35zer720j4w383xazxnmg09l6d8krcgvskmfhlfhvmxlj3c2cfrf99snaddr
naddr1qqtk5atdvfkx2amfwdcz6arjw4ehgttnvdhhyetnqyv8wumn8ghj7un9d3shjtn8w4k82em4d36jumt0v5pzqe7zxsty0ef7246y7ym5g60dpuhlf57c0ppjzmdxlaxanvm728ptqvzqqqr4gu9s27mkKind-30023 (Article)
Your Web of Trust
When you log in, jumblewisp builds a Web of Trust (WoT) from your social graph:
- 1 hop: everyone you follow
- 2 hops: everyone your follows follow
This gives you a pool of typically thousands of pubkeys — your extended network.
Two Signals
For any given account, we track two things:
- Follows — how many of your direct follows also follow them
- Mutes — how many people in your WoT have muted them
The Formula
We use Bayesian-smoothed exponential decay:
followRate = follows / myFollowSetSize muteRate = mutes / wotSize priorRate = 2%
smoothedRatio = (muteRate + priorRate × 10%) / (followRate + priorRate) score = round(100 × exp(−5 × smoothedRatio))
Dividing by myFollowSetSize means signal strength scales with your curation habits. If you follow 100 people, a single follow carries real weight. If you follow 10,000, it contributes far less — because mass-following is a weaker endorsement.
The prior keeps scores stable for accounts with little data. The exponential decay means even a small mute-to-follow ratio drops the score sharply.
What the Numbers Mean
| Situation | Score | | ------------------------------- | ----- | | Well-followed, never muted | ~87 | | Unknown to your network | 1 | | Muted by many, followed by none | 0 |
A score of 1 means the account is simply unknown — no one in your network follows or mutes them. A score of 0 means your WoT has actively flagged them with no countervailing follows.
Why This Approach
Your follow and mute graph is already a distributed reputation system — jumblewisp just does the math. No central authority decides who's trustworthy. Your network does.
Raw JSON
{
"kind": 30023,
"id": "3054a6c465a19ea3c1e76c0b9e240275975e8e863ee2e608650bd45d13038b1a",
"pubkey": "67c2341647e53e55744f1374469ed0f2ff4d3d87843216da6ff4dd9b37e51c2b",
"created_at": 1776950062,
"tags": [
[
"d",
"jumblewisp-trust-scores"
],
[
"title",
"How jumblewisp scores trust"
],
[
"summary",
"jumblewisp shows you a trust score (0–100) next to every account. Here's how it works."
],
[
"published_at",
"1776949989"
],
[
"client",
"nostria"
]
],
"content": "## Your Web of Trust\nWhen you log in, jumblewisp builds a **Web of Trust (WoT)** from your social graph:\n\n- **1 hop:** everyone you follow\n- **2 hops:** everyone your follows follow\n\nThis gives you a pool of typically thousands of pubkeys — your extended network.\n## Two Signals\nFor any given account, we track two things:\n\n- **Follows** — how many of your direct follows also follow them\n- **Mutes** — how many people in your WoT have muted them\n## The Formula\nWe use Bayesian-smoothed exponential decay:\n\n```\nfollowRate = follows / myFollowSetSize muteRate = mutes / wotSize priorRate = 2%\n\nsmoothedRatio = (muteRate + priorRate × 10%) / (followRate + priorRate) score = round(100 × exp(−5 × smoothedRatio))\n```\n\nDividing by `myFollowSetSize` means signal strength scales with your curation habits. If you follow 100 people, a single follow carries real weight. If you follow 10,000, it contributes far less — because mass-following is a weaker endorsement.\n\nThe prior keeps scores stable for accounts with little data. The exponential decay means even a small mute-to-follow ratio drops the score sharply.\n## What the Numbers Mean\n\n| Situation | Score |\n| ------------------------------- | ----- |\n| Well-followed, never muted | ~87 |\n| Unknown to your network | 1 |\n| Muted by many, followed by none | 0 |\n\nA score of **1** means the account is simply unknown — no one in your network follows or mutes them. A score of **0** means your WoT has actively flagged them with no countervailing follows.\n## Why This Approach\nYour follow and mute graph is already a distributed reputation system — jumblewisp just does the math. No central authority decides who's trustworthy. Your network does.",
"sig": "92e34658426e5996908ea150b6a99dbd064301730dd65da1327d574893820d7bf18e78b8bde1999ccb0fbc869b5f980c94acffe28b6661e3ed3e113665f52962"
}