Independent browser-local gate · ARD v0.91 Proposal checked August 27, 2026

Is your agent catalog ready for ARD?

Review entry quality, publishing, search, federation, relevance, trust, curation, runtime handoff, monitoring, and revocation without connecting to a registry.

Catalog evidence

Check only current implementation, conformance, policy, and failure-test evidence.

The ARD proposal version, schemas, conformance tooling, publisher and consumer implementations, feature flags, owners, support, and migration plan are pinned.Every resource has an approved stable identifier, publisher domain, artifact type, exactly one URL or data value, representative queries, capabilities, version, and source.Static discovery paths, JSON-LD, rel links, optional predecessor compatibility, crawling, namespaces, ingestion, freshness, deletion, and duplicate handling are tested.Required search plus any optional explore, list, federation, referral, wrapper, filter, pagination, relevance, cache, timeout, and partial-result behavior is validated.Collection scope, publisher admission, license, vulnerability, provenance, trust framework, signature verification, expiry, deprecation, and revocation policies are approved.Relevance judgments, exact lookup, ambiguous and adversarial queries, rare resources, keyword stuffing, stale entries, duplicates, and trust-score separation are evaluated.Install/connect approval, artifact validation, runtime identity, authentication, data scope, action authority, human confirmation, budget, logs, and cancellation remain separate from discovery.Monitoring covers entry and artifact drift, endpoint health, trust changes, runtime incidents, federation peers, revocation, cache purge, rollback, export, and recovery.

Local result

Complete the evidence review

Eight gates remain. Pin the proposal version and validate entry, collection, and trust policy before joining discovery to runtime.

A passing checklist does not prove conformance, trust, safety, interoperability, approval, or runtime authorization. Start with a curated non-production collection.

Continue by decision

v0.91 specificationEntries, JSON-LD, static publishing, REST search, trust, and conformance. Federation and rankingCollections, relevance, publisher identity, trust, referrals, approval, and drift. Choose the layerSeparate ARD, registries, MCP, A2A, authentication, invocation, and policy.

ARD catalog questions

No. It runs locally in the browser and does not publish a manifest, crawl a domain, validate an entry, search a registry, verify a trust manifest, install a resource, connect MCP, invoke A2A, or change policy.

No. Use the official schema, OpenAPI contract, and conformance tool for the pinned proposal version. This checklist covers additional operational and governance evidence.

No. ARD v0.91 says the score is semantic relevance only and must not be interpreted as cryptographic trust, compliance, or safety.

No. ARD discovers descriptions. The artifact protocol and your runtime handle authentication, data and action authorization, approval, monitoring, cancellation, and recovery.

The specification permits broad collections, but the ARD project expects most enterprises to use governed approved sets. Start curated and expand only with tested ingestion, trust, and runtime controls.

Official references: ARD v0.91 proposal, how ARD works, and AWS federation context.

Is your agent catalog ready for ARD?

Catalog evidence

Complete the evidence review

Continue by decision

ARD catalog questions

Features

Resources

Company