Lab interoperability programme

This programme describes how ARPA should collect interoperability evidence before any v1.0 or broader lab proposal claims are made.

Purpose

Interoperability is not demonstrated by one reference implementation or by schema validation alone. The lab should collect evidence that independently built or independently operated systems can exchange registry-relevant state while preserving ARPA semantics.

Interoperability dimensions

Dimension Question
Syntax Can systems parse the same records, events, receipts and evidence bundles?
Semantics Do systems reach the same authority, lifecycle and historical-resolution outcomes?
Negative behavior Do systems reject the same invalid, stale, conflicted or unsupported inputs?
Federation Can state be projected across registries without laundering authority?
Discovery Can systems expose findability without implying authorization?
Evidence Can reviewers reconstruct why a decision was reached?
Operations Can operators revoke, suspend, transfer and recover without silent ambiguity?

Evidence levels

Level Description
L0 — Repository-local Repository-owned implementation and vectors only
L1 — Cross-runtime More than one implementation track under repository stewardship
L2 — External codebase Separate codebase consumes ARPA artifacts
L3 — Independently operated Distinct operator runs tests and reports evidence
L4 — Multi-party exercise Multiple operators exchange records or events in a bounded plugfest

Current repository-owned Python and TypeScript tracks are valuable L1 evidence. They should not be overstated as L3 or L4 evidence.

Candidate plugfest scenarios

  1. Read-only registry consumption — external client resolves agent identity, lifecycle state and evidence references.
  2. Delegated action resolution — two implementations evaluate the same delegated authority bundle.
  3. Federated projection — registry B consumes selected records from registry A with provenance and competence limits.
  4. Historical reconstruction — implementations reconstruct requested-time state from retained evidence and later material events.
  5. Discovery boundary — search or discovery output is consumed without being treated as authority.
  6. Revocation propagation — systems distinguish effective revocation from enforcement convergence.

Minimum test matrix

Scenario Positive case Negative case
Identity resolution Valid agentreg: identifier resolves Non-canonical or unsupported identifier fails safely
Lifecycle Active state permits reliance within scope Suspended or retired state blocks reliance
Delegation Scoped delegation is sufficient Delegation exceeds subject, action, time or issuer competence
Evidence Required proof inputs are present Missing or stale proof inputs fail closed
Federation Provenance and competence are preserved Imported record is treated as local authority
Historical Requested-time state is reconstructed with quality disclosure Later material event is hidden or ignored

Reporting

Interoperability reports should disclose:

  • implementation identity and operator;
  • ARPA baseline and profiles used;
  • fixtures and vectors executed;
  • deviations and unsupported features;
  • whether tests were run by repository maintainers, implementers or independent operators;
  • retained evidence bundle location.

Promotion path

Interoperability findings can become:

  • conformance vectors;
  • implementation guide updates;
  • protocol clarification amendments;
  • experiment notes;
  • v1.0 readiness evidence;
  • lab proposal supporting material.

Boundary

This programme does not claim completed interoperability. It defines how the repository will collect and classify evidence so future claims are appropriately scoped.


Agent Registry Protocol documentation. Specification text and documentation are licensed as stated in the repository NOTICE and license files.

This site uses Just the Docs, a documentation theme for Jekyll.