RFC-038: Domain of Concern Registry — AIGP Dialects as First-Class Artifacts — 11. Registry Governance
AIGP Specification › RFC-038: Domain of Concern Registry — AIGP Dialects as First-Class Artifacts › 11. Registry Governance
← 10. The Scientific Parallel: Dialects as Research Protocols · Section index · 12. Distribution Model →
11. Registry Governance
11.1 Registry Authority
A Dialect Registry MAY be operated by:
- The AIGP protocol authority (canonical registry)
- An industry consortium (sector-specific registry)
- A regulatory body (jurisdiction-specific registry)
- An organization (private registry for internal dialects)
Multiple registries MAY coexist. A governed application MAY subscribe to dialects from different registries. Registry trust is established through the same HMAC and D-DNA mechanisms used throughout AIGP.
11.2 Publisher Accreditation
Not every entity may publish to a registry. A publisher MUST demonstrate:
| Requirement | Rationale |
|---|---|
| Domain expertise in the concern class | Prevents uninformed dialect construction |
| Measurement science literacy (construct validity, Stevens scales) | Prevents malformed variable declarations |
| Governance infrastructure to maintain the dialect | Prevents abandoned dialects without successors |
| Declared conflict-of-interest statement | Prevents dialects designed to produce favorable verdicts |
The registry operator determines publisher accreditation criteria. The canonical AIGP registry SHOULD require peer review of new dialects before publication.
11.3 Dialect Review
Before a dialect is published to the canonical registry, it SHOULD undergo structured review:
| Review Dimension | Question |
|---|---|
| Construct validity | Do the variables actually represent the declared concern? |
| Measurement validity | Are types, scales, and gap operators consistent? |
| Evidence grounding | Are admissibility levels appropriate for the concern severity? |
| Completeness | Are there obvious concern indicators missing from the model? |
| Overfit | Does the dialect measure too many things, diluting signal? |
| Known invalidity | Are the known_invalid_uses declarations honest and complete? |
| Warrant quality | Are the Toulmin warrants for each variable defensible? |
| Threshold calibration | Are default thresholds empirically grounded or arbitrary? |
A dialect that passes review is published. A dialect that fails review receives structured feedback and may be resubmitted.
11.4 Registry Integrity
The registry MUST maintain:
- Immutability — Published versions cannot be modified
- Availability — Subscribed applications can resolve dialects at CHECK time
- Auditability — All publish, deprecate, and subscribe operations are logged
- Tamper evidence — Bundle hashes are verified on every resolution
- Lineage — The full version history of every dialect is preserved
← 10. The Scientific Parallel: Dialects as Research Protocols · Section index · 12. Distribution Model →