Operational observation evidence

Experimental, optional profile for issue #97. Not part of the normative ARPA v0.10.0 Candidate Specification or the published IETF draft.

ARPA’s Service Endpoint and Status records already describe the operational state. The observation evidence schema instead describes a particular observer’s reported measurement of a particular endpoint. It is not a new authoritative record family. A self-reported or independently reported observation does not by itself establish observer competence, recognition, assurance, action permission, current operational readiness or future availability.

Evidence processing flow

flowchart TD
  A["Choose endpoint and observation method"] --> B{"Probe attempted?"}
  B -->|No| C["not_performed"]
  B -->|Yes| D{"Supported outcome?"}
  D -->|Success| E["success"]
  D -->|Failure| F["failure + reason"]
  D -->|Unclear| G["indeterminate"]
  C --> H["Evidence claim + observer + observation time"]
  E --> H
  F --> H
  G --> H
  H --> I["Schema validation"]
  I --> J["Optional independent evidence verification"]
  J --> K["Derived evidence consumer"]
  K -. "No implied status/authority mutation" .-> L["Separate status and authority evaluation"]

Single-attempt state diagram

This state diagram describes an observation attempt, not an ARPA agent lifecycle or status transition. Each subsequent measurement creates a separate observation.

stateDiagram-v2
  [*] --> Attempt
  Attempt --> NotPerformed: no probe
  Attempt --> Success: defined successful outcome
  Attempt --> Failure: defined failed outcome
  Attempt --> Indeterminate: insufficient evidence
  NotPerformed --> [*]
  Success --> [*]
  Failure --> [*]
  Indeterminate --> [*]

Interpretation boundaries

Method What success can establish What it cannot establish
transport_probe Transport condition held at observation time A2A protocol validity
protocol_probe Declared protocol interaction succeeded Task authority, task completion
endpoint_control_proof Declared endpoint control procedure succeeded Institutional recognition, delegation
other_declared_method Only declared method’s tested condition Any stronger untested claim

An observation is a provenance-bound claim. Evidence digests bind referenced bytes only if independently checked. Schema validation does not authenticate the observer or cryptographically verify its evidence. failure from one vantage point is not necessarily authoritative service unavailability. indeterminate and not_performed must not be counted as success.

observed_at is distinct from the optional issued_at. Historical evidence remains usable for permitted historical analysis without asserting current state. Repeated reports must not be counted as independent successful observations without appropriate deduplication and evidence verification. disclosure_class is advisory metadata; actual access restrictions belong at the host’s enforcement boundary. Minimize correlated endpoint/observer identifiers and infrastructure disclosures.

Validate

Run python3 scripts/validate_operational_observations.py for the positive and negative fixtures. Use --output artifacts/operational-observation/validation-report.json to generate a reproducible fixture report. This tests schema and expected invalid representations only; it is not proof of live operational availability, independent security review or interoperability. Future #98 may aggregate observations; #100 will test composition with ARPA reliance evaluation.


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.