Trust Graph Artifacts agentic-governance alignment

The Trust Graph Artifacts (TGA) Agentic Systems Architecture and Governance guide describes a protocol-neutral architecture for bounded jobs, legitimate authority, delegation, capability separation, governed execution, evidence, revocation, remediation, and assurance.

ARPA is a concrete realization of selected discovery, authority-control, lifecycle, federation, and evidence responsibilities in that architecture. This relationship is informative and optional: adopting the TGA guide does not require ARPA, and ARPA does not depend on TGA for protocol authority.

Authority boundary

TSMM  -> canonical trust-system semantics
TGA   -> generic agentic governance patterns and assurance guidance
TIS   -> portable executable contracts
ARPA  -> agent-registry and authority-control protocol realization

ARPA remains authoritative for its modules, normative requirements, controlled vocabulary, conformance profiles, API/event behavior, and implementation evidence. TGA remains authoritative for its guide, patterns, assurance cases, and guide-level negative-test methodology.

Four-plane placement

Plane	TGA architectural responsibility	ARPA contribution
Discovery	locate a persistent role/agent without treating discovery as permission	ARPA-Core registration, lifecycle, discovery and historical resolution
Authority control	determine whether a bounded effect is legitimate now	ARPA-Relations + ARPA-Authority
Execution	perform an admitted job through constrained capabilities and enforcement	outside ARPA’s general responsibility; ARPA may supply authority/evidence inputs
Evidence and assurance	prove authority, decisions and consequential action are reconstructable	ARPA-Assurance + ARPA-Evidence

ARPA-Federation contributes governed recognition and withdrawal across registry domains; technical federation alone does not create governance recognition.

Shared architectural invariants

An ARPA implementation used within the TGA guide must preserve at least these rules:

1. identity does not imply authority;
2. registration or discovery does not imply permission to act;
3. capability does not imply permission;
4. a relationship does not imply delegated authority;
5. delegated scope may narrow but not silently expand;
6. revocation is incomplete until relevant enforcement surfaces converge;
7. technical federation does not imply governance recognition;
8. consequential action must remain reconstructable;
9. successful execution does not prove a legitimate effect.

Module mapping

Guide concern	ARPA module(s)
persistent agent identity and lifecycle	ARPA-Core
principal/operator relationships	ARPA-Relations
scoped capability assurance	ARPA-Assurance
delegation and authority evaluation	ARPA-Authority
receipts, reconstruction and audit	ARPA-Evidence
cross-domain recognition	ARPA-Federation

Responsibilities intentionally outside ARPA

The TGA guide also covers concerns that are not ARPA protocol responsibilities, including job decomposition, model/runtime orchestration, capability brokering, policy-enforcement implementation, fan-out convergence, generic effect-admission architecture, and TIS portable-contract selection.

This prevents the registry/control plane from being mistaken for the complete agentic runtime.

Machine-readable declaration

The repository declaration is:

interop/tga-agentic-governance-alignment.json

It is checked by scripts/validate_tga_alignment.py as part of make validate and therefore make release-check-all.

The reciprocal TGA declaration is maintained by TGA under bindings/arpa/tga-arpa-agentic-governance-alignment.json.

Neither declaration constitutes external certification or cross-repository conformance.