Skip to content

Governance and Risks

Source version 1.2.0. Draft for discussion and validation.

17. Governance model

17.1 Governance principles

  • One authoritative meaning: Serialisations and interfaces must derive from the same governed semantic source.
  • Profiles before platforms: Agree the publication contract before selecting or extending infrastructure.
  • No hidden transformations: Every generated representation must be traceable to its source and transformation activity.
  • Version everything independently: Artefact, profile, vocabulary, schema, mapping and interface versions must be explicit.
  • Persistent metadata: Withdrawn resources retain resolvable metadata and status.
  • Executable conformance: Narrative requirements should be backed by validation rules wherever practical.
  • Federation-aware design: Global semantics remain consistent while MOs can publish conformant local distributions and services.
  • Open standards first: Reuse W3C, OpenAPI Initiative and recognised agent protocol specifications before adding GS1-specific extensions.

17.2 Candidate decision rights

Decision Proposed owner or forum
Authoritative standards content Relevant GSMP governance and 8.3.a business owners
Core semantic publication profile 8.3.b with Standards, Web Technology and Product review
GS1 semantic vocabulary terms Established semantic governance authority
OpenAPI agentic overlay Joint 8.3.b and 8.4.b review
Runtime interface behaviour 8.4.b and product/service owner
FAIR conformance criteria 8.3.b with FAIR and data governance expertise
Publication platform implementation Relevant product and technology owner
External protocol alignment Architecture, Standards and ecosystem liaison governance

18. Key risks and controls

Risk Consequence Control
Format-first delivery JSON or Markdown is labelled AI-ready without semantic governance. Require profile conformance and machine-readable evidence.
Duplicate semantics OpenAPI, MCP and A2A descriptions diverge. Generate projections from a shared semantic source and test equivalence.
Profile proliferation Every project creates a bespoke profile. Use a core profile plus reusable overlays.
Unstable identifiers Citations, mappings and agent references break. Approve persistent identifier and tombstone policy early.
Lossy Markdown conversion Normative meaning or table structure is lost. Use source-preserving transformation and structured sidecars.
FAIR interpreted as open Restricted or sensitive data is exposed incorrectly. Separate FAIR accessibility from open access; publish auth and policy metadata.
OpenAPI extension sprawl Tooling ignores uncontrolled x- fields. Prefer standard fields and Overlay documents; govern a minimal extension registry.
Semantic context drift OpenAPI schemas and JSON-LD contexts evolve independently. Version them independently, bind them through profile metadata and run schema-to-context equivalence tests.
Unsafe remote context resolution A compromised or unexpected JSON-LD context changes payload meaning or triggers unsafe retrieval. Use versioned immutable contexts, integrity checks, caching and an approved context allow-list.
Automatic over-exposure Every OpenAPI operation is generated as an MCP tool regardless of risk or relevance. Require explicit operation eligibility and human/security approval before projection.
Agent unsafe action An agent invokes a state-changing operation without confirmation. Declare side effects, idempotency, risk and confirmation requirements.
Semantic drift Artefacts and mappings change independently. Version dependencies and automate impact assessment.
Platform scope creep 8.3.b becomes a broad implementation programme. Keep scope on profiles, publication pipelines, validation and initial pilots.