Independent implementation guide
This guide explains how an external implementer can participate in the ARPA lab without adopting the whole repository or claiming premature conformance.
Goal
The goal is to make independent implementation evidence cheap enough to produce and precise enough to be useful. An implementation report should help reviewers distinguish:
- implemented behavior;
- intentionally unsupported behavior;
- profile-specific behavior;
- deviations from the Candidate Specification;
- ambiguous requirements that need clarification.
Minimum useful implementation
A minimum useful implementation should support at least one of the following paths:
| Path | Description |
|---|---|
| Read-only registry | Publishes agent identity, lifecycle state and evidence references without write APIs |
| Discovery registry | Supports controlled discovery while preserving discovery-is-not-authority semantics |
| Authority registry | Resolves authority, delegation and lifecycle state for a bounded domain |
| Federated projection | Consumes or projects selected state from another registry with provenance and competence limits |
| Historical resolver | Reconstructs requested-time state from retained evidence and later material events |
An implementer does not need to implement every ARPA module to provide useful evidence. The implementation report must disclose the supported profile and known limits.
Evidence package
A useful external report should contain:
- implementation name, version and operator;
- implementation type and supported profile;
- source repository or artifact reference if available;
- schema version and normative baseline used;
- test vector results;
- positive cases passed;
- negative cases rejected;
- known deviations;
- unsupported features;
- security, key and policy assumptions;
- evidence retention behavior;
- statement on whether the implementation was operated independently from this repository.
Suggested report structure
# ARPA implementation report
## Implementation identity
## Operator and independence statement
## Supported profile
## Normative baseline
## Test environment
## Positive vector results
## Negative vector results
## Historical-resolution behavior
## Federation or projection behavior
## Known deviations
## Unsupported features
## Evidence bundle
## Security and privacy notes
Independence language
Use precise independence language. For example:
| Claim | Appropriate when |
|---|---|
| Repository-owned implementation track | The code is maintained in this repository or by the same steward |
| External implementation | The codebase is maintained elsewhere but may still be influenced by ARPA contributors |
| Independently operated implementation | The implementation was deployed, run and reported by a distinct operator |
| Independent conformance evidence | The report and evidence were produced without relying on repository-controlled execution |
The repository-owned Python and TypeScript tracks improve implementation diversity, but they are not external organizational independence evidence.
Submission path
Use the repository’s implementation report issue template when submitting external evidence. Include links to artifacts, reproducibility notes and any cases where ARPA requirements were unclear.
What not to claim
Do not claim:
- production certification;
- legal authority;
- universal interoperability;
- security review completion;
- external independence;
- profile conformance;
unless the submitted evidence directly supports that claim.
Lab value
Independent implementation reports are useful even when they fail. Failures reveal ambiguous requirements, unrealistic assumptions, missing vectors and hidden coupling in the reference implementation.