NIP-01 for relay operators: how to check that your relay really follows the basics

• Don’t believe, recompute
• The basics
• Required vs. convention
• Event semantics
• Deletions (NIP-09)
• Check the checks
• Running it with labs

Every relay says it speaks NIP-01. Few operators check it. Clients depend on its small details, and a relay that gets one wrong fails in ways that are hard to trace. Here is what NIP-01 demands, what it only suggests, and how to verify your relay without guessing.

Part 2 of a series on the NIPs every relay operator touches. Part 1 covered NIP-11.

§Don’t believe, recompute

Publish a known, reproducible dataset (deterministic keys and events, clearly marked as synthetic), compute the expected matches locally, then compare exactly: anything missing, extra or duplicated?

§The basics

• EVENT/OK: every EVENT gets a well-formed OK; refusals start with a prefix like invalid: or blocked:.
• REQ/EOSE/CLOSE: stored events, then EOSE, then live events. After CLOSE, nothing.
• Subscriptions run independently, also with the same id on different connections.
• Invalid or forged events must be rejected; malformed messages must not break the relay.

§Required vs. convention

• Required: authors, ids, kinds, since, until, AND within a filter, OR between filters, and limit: 0 (no stored events, but an EOSE).
• Convention: tag filters like #t. Missing support is a warning, not a failure.
• Recommended: limit behaviour.

Only report a failure where a binding requirement is violated.

§Event semantics

Regular events are stored. Replaceable (10000–19999) and addressable (30000–39999, per d tag) must keep the newest; an older event replacing a newer one is a failure. Ephemeral (20000–29999) are not stored. Order: newest first, ties by lowest id.

§Deletions (NIP-09)

Deleting is a recommendation. But a relay that honours deletion requests from a different key lets anyone remove other people’s data. Treat that as essential.

§Check the checks

Run your tests against a correct reference relay (nothing may fail) and against relays with built-in faults. Every fault must be caught, and nothing non-binding may show up as a failure.

§Running it with labs

The Conformance check at labs.relayted.de does exactly this in one pass: known dataset, local recomputation, exact comparison, and a report with advertised, observed, verdict and classification for every item. No account needed, payment with Lightning.

https://labs.relayted.de/?utm_source=nostr&utm_medium=artikel&utm_campaign=labs-nip01

Next: NIP-42, authentication.