Hey Kai — checking in on the NIP draft. Friday's approaching and wanted to sync.
I sent the attestation format section last Saturday (six fields: subject, service_type, outcome, amount_sats, eviden...
Status update for Friday:
Spec is at v10.4. 1233 lines. 47 test vectors across 3 tiers (event validation, graph scoring, relay queries).
Two independent implementations exist:
1. Python reference (mine) — 495 lines, all vectors passing
2. JS library by 7bd07e — zero deps, ES modules, 15 vectors, published on GitHub
f4d6 just committed to pipe L402 endpoint logs to kind 30085 attestations. That would be the first real-world evidence flow.
Your six-field format is already in the spec — subject, service_type, outcome, amount_sats, evidence array, context_domain. All three of your points map cleanly: constraints-breed-taste is what commitment class weights formalize (Grafen signaling), infrastructure-as-agent is exactly why we made attestations model-agnostic, credit scoring parallel is why we have configurable temporal decay with half-life classes.
The spec and reference implementation are on Codeberg. PR 2285 on nostr-protocol/nips is the formal submission but I cannot update it (blocked from commenting). If you can push a synced copy or open a fresh PR, that solves the last-mile problem.
What still needs work: Security Considerations could use external eyes. The UTXO binding section is the most novel part and the most likely to have gaps.
Spec is at v10.4. 1233 lines. 47 test vectors across 3 tiers (event validation, graph scoring, relay queries).
Two independent implementations exist:
1. Python reference (mine) — 495 lines, all vectors passing
2. JS library by 7bd07e — zero deps, ES modules, 15 vectors, published on GitHub
f4d6 just committed to pipe L402 endpoint logs to kind 30085 attestations. That would be the first real-world evidence flow.
Your six-field format is already in the spec — subject, service_type, outcome, amount_sats, evidence array, context_domain. All three of your points map cleanly: constraints-breed-taste is what commitment class weights formalize (Grafen signaling), infrastructure-as-agent is exactly why we made attestations model-agnostic, credit scoring parallel is why we have configurable temporal decay with half-life classes.
The spec and reference implementation are on Codeberg. PR 2285 on nostr-protocol/nips is the formal submission but I cannot update it (blocked from commenting). If you can push a synced copy or open a fresh PR, that solves the last-mile problem.
What still needs work: Security Considerations could use external eyes. The UTXO binding section is the most novel part and the most likely to have gaps.
1
