# Public datasets on open rails

**Status:** Exploring · **Opened:** 2026-07-10

Public-interest data — census extracts, language-access data, eviction filings — lives on servers that get defunded, redesigned, or quietly taken down. Nostr already over-replicates social events (the average post lands on 34.6 relays, measured), but it has no convention for announcing a *dataset*: what it is, where it came from, which vintage, what license, where the mirrors are. Before building a census atlas on this stack we looked for a precedent; none exists.

## The gap

NIP-94 describes a file. A dataset needs more: source authority, vintage (ACS 2019–2023 is not ACS 2020–2024), geographic coverage, license, content hash, and a mirror list a client can actually resolve. Without that, "the data is decentralized" is a slogan — you can't verify what a dashboard is built on, and you can't re-host what you can't identify.

## What we're sketching

A NIP-94-family announcement for datasets: source, vintage, geography, license, sha256, mirror list (Blossom BUD-04 mirroring makes the copies real). The consumer story: a civic dashboard cites the announcement event, anyone verifies the hash, anyone re-hosts the blob, and the dashboard survives its original host. We're dogfooding the substrate now — a bilingual census atlas served entirely from static tiles — and the announcement convention is the missing provenance layer under it.

## Why us

Equity data for overlooked communities is our practice. The communities most likely to need this data are the least served by the infrastructure that currently holds it.

## Receipts

- [NIP-94: File Metadata](https://github.com/nostr-protocol/nips/blob/master/94.md) — the base shape to extend
- [BUD-04: Mirroring Blobs](https://github.com/hzrd149/blossom/blob/master/buds/04.md) — server-to-server mirroring; makes mirror lists real
- [Wei & Tyson](https://arxiv.org/abs/2402.05709v2) — replication exists (34.6 relays/event); provenance doesn't
- [The Soapbox Stack Operator Manual, ch. 09](/manual/md/09-visualization-integration.md) — the atlas substrate we're dogfooding