block·health

A single-page, no-build site — a mempool.space-style block row that shows the newest Bitcoin blocks as they arrive over NIP-333 (Nostr kind 33333), validates them in the browser, and scores each block for OP_RETURN datacarrier health.

Everything is static: open index.html from any host (GitHub Pages, a pod, file:// via a local server). No backend, no build step, no framework.

What it does

1. Fetches the live header stream — pulls the newest kind-33333 event (mainnet d=btc) from several Nostr relays, verifies the event's NIP-01 id hash and BIP-340 signature, and takes the newest valid one. The publisher key is a filter, not an authority.
2. Verifies the 12 headers — checks each header's proof-of-work and its linkage to the previous header. A lying relay or publisher is detected here, not trusted.
3. Validates each full block (no prevouts) — fetches the raw block from esplora (mempool.space, falling back to blockstream.info), decodes it with the bitcoin-kernel engine in a Web Worker, and runs the structure and block-context rulesets. Prevout resolution is deliberately skipped — this is a pruned window with an empty coin view — so value/fee rules honestly report "unresolved" while merkle root, sizes, weight, sigops and the witness commitment are fully checked.
4. Scores datacarrier health — for every OP_RETURN output it measures the pushed data size and flags outputs over the historical 80-byte data / 83-byte scriptPubKey standard. The coinbase witness-commitment OP_RETURN is counted separately as protocol overhead, never as user data. Click any block for the full breakdown.

Layout

index.html            the whole app (styles + logic inline, ES module)
validator-worker.js   off-main-thread: fetch raw block, decode, validate, measure health
engine/               vendored bitcoin-kernel engine (codec + JSON-LD schemas)

The engine/ directory is copied verbatim from bitcoin-kernel/browser-node so the site is fully self-contained on a static host.

Trust model

Bitcoin headers are self-certifying (proof-of-work + linkage), so no relay, publisher, or esplora host is trusted. Consumers re-check everything on receipt; redundancy is additive. See the NIP-333 spec for the header-distribution details.

Run locally

python3 -m http.server 8899   # then open http://localhost:8899/

A server (not file://) is needed because the page loads ES modules and a module worker.