Public Hyperliquid (HIP-4) prediction-market catalog snapshots and a permanent settlement archive. This is the data plane behind purrbook.xyz's crawl surface — the site renders from these files whenever its live indexer is unreachable, so the market pages, sitemap, and llms.txt never go dark.
| File | What it is | Update behavior |
|---|---|---|
catalog.json |
Every live HIP-4 question and outcome: names, verbatim resolution rules, YES-side mid prices with per-run timestamps | Overwritten by the pipeline (~10-minute cadence) |
archive.json |
Permanent record of settled outcomes: identity, resolution rule, last observed price, and when settlement was observed | Append-only — entries are never rewritten or removed |
meta/overlay.json |
Display curation (event groupings, competition labels) mirrored from the Purrbook indexer | On curation changes |
Fetch raw from main — no auth, no rate ceremony:
https://raw.githubusercontent.com/0xm0w/purrbook-data/main/catalog.json
https://raw.githubusercontent.com/0xm0w/purrbook-data/main/archive.json
Raw CDN propagation from commit is typically seconds. generatedAt and every
price's priceAt tell you exactly how fresh what you're holding is.
- Every price carries its observation timestamp — nothing is presented fresher than it is.
- Resolution rules are the venue's text, verbatim — never paraphrased. Markets whose on-chain description is machine metadata carry an empty rule rather than a fabricated one.
- Archive entries never claim a winner — only observed prices and the rule. Settlement outcomes belong to the venue's resolution, not to inference.
- Non-tradable fallback legs are flagged (
isFallback) and never archived.
A zero-dependency Node script (snapshot.mjs / run.mjs) executes on a
GitHub Actions schedule, reading Hyperliquid's public info API and committing
diffs as purrbook-snapshot. When an outcome leaves the live catalog, it is
frozen into the archive in the same run — settlements are captured the moment
they happen, unattended. A small Cloudflare Worker (tools/watchdog/)
re-dispatches the workflow if GitHub's scheduler drifts.
The pipeline owns main. Human commits are limited to the writer/tooling code
via PR; the data files are machine-written.
Code is MIT. The JSON data files record public Hyperliquid market facts.