EXP-001 — Federated registry conflict
This experiment packet is informative. It does not modify the ARPA Candidate Specification, the IETF Internet-Draft track, conformance profiles, schemas, or implementation behavior.
Hypothesis
Two registries can exchange, project, or reference agent state without implying that the receiving registry accepts the source registry’s authority for every relying context.
A useful federation model should let a relying party distinguish:
- an authority conflict;
- a projection conflict;
- a freshness failure;
- a relying-party policy conflict;
- an indeterminate reconstruction boundary.
Scenario
A relying party receives information about the same agent from two registries.
- Registry A is competent for the agent’s original registration and lifecycle state.
- Registry B publishes a projection used by a relying workflow.
- Registry A records a suspension event.
- Registry B still exposes an active projection because of delay, policy difference, cache staleness, or incomplete event propagation.
- The relying party asks whether the agent may be accepted for a consequential workflow.
The experiment tests whether the registry and evidence model can represent the conflict without falsely treating federation as transitive trust.
Participants and roles
| Participant | Role |
|---|---|
| Source registry | Publishes original agent state, lifecycle events, delegation references, and retained evidence. |
| Receiving registry | Publishes a projection, mirror, cache, summary, or federation-derived view. |
| Agent controller | Has a relationship to the agent and may trigger lifecycle or delegation updates. |
| Relying party | Evaluates whether the available evidence is acceptable for a workflow. |
| Auditor or reviewer | Examines retained records and determines whether the experiment result is reproducible. |
Input conditions
The experiment should define fixture data for at least four paths:
| Case | Source state | Receiving state | Expected classification |
|---|---|---|---|
| A | Active | Active | No conflict detected, subject to policy acceptance. |
| B | Suspended | Active | Freshness failure or projection conflict. |
| C | Revoked | Active | Authority conflict unless receiving registry has a competent override. |
| D | Active | Suspended | Relying-party policy conflict or receiving-registry local restriction. |
| E | Missing retained records | Active | Indeterminate reconstruction boundary. |
Expected behavior
A successful experiment should show that an implementation or evidence packet can:
- identify the source of each claim;
- identify which source is competent for which claim type;
- distinguish original authority from projection;
- disclose freshness, cache, retention, and event-sequence limits;
- avoid silently upgrading a projection into authority;
- produce a relying-party decision receipt that states the basis of acceptance, rejection, or indeterminacy.
Evidence to retain
The experiment packet should retain or produce:
- source registry fixture records;
- receiving registry projection records;
- lifecycle event records;
- relevant policy or competence declarations;
- evaluation receipt;
- conflict classification;
- freshness and cache metadata;
- retained-record provenance;
- implementation notes and known deviations.
Outcome model
| Outcome | Meaning |
|---|---|
| Pass | The packet distinguishes authority, projection, freshness, and relying-policy conflicts using retained evidence. |
| Fail | The packet collapses federation into transitive trust or presents stale projected state as authoritative state. |
| Indeterminate | Required retained records, policy checkpoints, or competence declarations are absent or contradictory. |
| Out of scope | The scenario depends on rules or claims outside the declared ARPA profile or experiment boundary. |
Unresolved questions
- What minimum metadata is needed to make projection freshness inspectable?
- Should competence be declared per registry, per claim type, per agent class, or per relying-party context?
- How should implementations expose cache-derived state without encouraging unsafe reliance?
- When should an experiment result become a conformance vector rather than remain a worked scenario?
Promotion path
If useful, results from this experiment may later be promoted into:
- negative conformance vectors for authority laundering;
- implementation-report requirements for federation and projection behavior;
- a registry-federation research note revision;
- a profile-specific clarification if repeated independent implementation evidence shows stable behavior.
Promotion requires a separate governed change. This experiment does not itself create normative requirements.
Related issues
- Closes #76 when merged.
- Supports #79.