Lab contribution pathways
The ARPA lab accepts more than code patches. Agent governance needs evidence from implementers, operators, reviewers, policy practitioners, security reviewers and affected communities.
This document complements ARPA’s existing protocol specification, IETF submission, implementation, conformance, governance and release material. It does not create a separate governance process and it does not change normative protocol requirements.
Contribution types
| Contribution | Useful output |
|---|---|
| Specification feedback | Ambiguity report, missing requirement, unsafe implication, terminology issue |
| Research question | Bounded question that may become a note, ADR, experiment or later proposal |
| Implementation report | Evidence from a repository-owned, external or independently operated implementation |
| Experiment proposal or result | Hypothesis, setup, observations, negative results and retained evidence |
| Interoperability report | Boundary mapping, implementation evidence, deviations and unresolved questions |
| Conformance vector | Positive or negative machine-checkable case |
| Deployment report | Pilot topology, operational assumptions and failure modes |
| Threat model | Attack path, control gap or adversarial test case |
| Privacy review | Correlation, minimization, retention or disclosure risk |
| Governance review | Accountability, redress, suspension, appeal or authority concern |
| Documentation improvement | Reader path, implementer guide or operator checklist improvement |
Preferred contribution framing
A strong contribution states:
- the decision or failure mode it addresses;
- the affected ARPA module, profile or guide;
- the evidence used;
- the proposed change or observation;
- the claim it does not make.
Good first lab contributions
- Run the quickstart and report any point where authority semantics are unclear.
- Implement a read-only registry profile and submit an implementation report.
- Add a negative vector for discovery being mistaken as authority.
- Compare ARPA lifecycle states with another agent framework’s status model.
- Write an experiment note for delegated authority revocation.
- Review evidence-retention assumptions from a privacy perspective.
- Test a registry-federation conflict and document the result.
Issue routing
| Contribution | Recommended issue type |
|---|---|
| Ambiguity or defect in protocol text | Specification feedback |
| External implementation evidence | Implementation report |
| Runtime or tool failure | Bug report |
| Bounded research question | Research question |
| Proposed lab-track experiment | Experiment proposal |
| Cross-runtime, cross-registry, adapter or external-system evidence | Interoperability report |
New intake templates
This repository provides structured issue forms for lab-track inputs:
- Research question — for bounded investigative questions that may lead to a note, ADR, experiment or later proposal.
- Experiment proposal — for hypotheses with setup, evidence, outcome criteria and promotion path.
- Interoperability report — for evidence from adapters, external implementations, profiles, projections or protocol-boundary tests.
These templates are intentionally scoped. Submitting one does not create a normative change, certification claim, or accepted conformance result.
Review principles
Contributions should be reviewed against these principles:
- preserve the separation of discovery, registration, authority, delegation, enforcement and assurance;
- prefer evidence over assertion;
- retain negative results when they prevent unsafe assumptions;
- avoid expanding ARPA into a universal AI-governance framework;
- disclose scope and independence precisely;
- keep normative changes gated by explicit specification process.
Community posture
The lab is useful only if it becomes legible to people outside the original authoring context. Contributions that improve external readability, reproducibility and implementability are high-value even when they do not add new protocol features.
Related issues
- Closes #78 when merged.
- Supports #79.