Motivation
The schema accepts architecture as a string or as a structured object with components and data flows, but the comprehensive example still uses a plain string — so the structured form is effectively undemonstrated. For real item-definition exchange across ISO 21434 and UN R155 stakeholders, the model also needs interfaces, attack-surface classification, environment variants, product hierarchy, trust zones, and safety relevance. Separately, assumptions are still free-text strings, which is too weak to support risk ownership and responsibility transfer between supplier and OEM.
Proposed change
architecture.interfaces[]: id, physical/logical type, protocol, direction (inbound|outbound|bidirectional), exposure (remote|adjacent|local|physical), trust zone reference.
architecture.trustZones[] and component.trustZoneId so zones become first-class with optional boundaryDescription and graphCoordinates (see related visualization issue).
component.attackSurface enum: remote | adjacent | local | physical | supplyChain | service | manufacturing.
itemDefinition.environment structured: markets[], vehiclePlatforms[], productionPhase, operationalPhase, serviceModes[].
- Product hierarchy:
component.kind enum (vehicle | system | ecu | softwareComponent | dataObject) and explicit parentComponentId semantics so a tree can be rendered.
- Safety relevance: optional
component.safetyRelevance with safetyGoalRef and asilContext (informational only — RDX does not assert ASIL ratings, only links them).
- Structured
assumption objects: { id, statement, owner, validityPeriod, responsibleParty, evidenceRefs[], riskSharingNote }. Keep a string-array form accepted for backwards compatibility, but mark deprecated.
- Rewrite the comprehensive example to exercise structured architecture and structured assumptions end-to-end.
References
spec/json/rdx.schema.json — architecture oneOf string|object
examples/rdx-xml-comprehensive.xml — still uses string architecture
- AI peer review, section 4 ("Architecture support exists, but is still too lightweight") and section 2 (structured assumptions)
Acceptance criteria
Surfaced by external AI peer review; see chat transcript for full review text.
Motivation
The schema accepts
architectureas a string or as a structured object with components and data flows, but the comprehensive example still uses a plain string — so the structured form is effectively undemonstrated. For real item-definition exchange across ISO 21434 and UN R155 stakeholders, the model also needs interfaces, attack-surface classification, environment variants, product hierarchy, trust zones, and safety relevance. Separately,assumptionsare still free-text strings, which is too weak to support risk ownership and responsibility transfer between supplier and OEM.Proposed change
architecture.interfaces[]: id, physical/logical type, protocol, direction (inbound|outbound|bidirectional), exposure (remote|adjacent|local|physical), trust zone reference.architecture.trustZones[]andcomponent.trustZoneIdso zones become first-class with optionalboundaryDescriptionandgraphCoordinates(see related visualization issue).component.attackSurfaceenum:remote|adjacent|local|physical|supplyChain|service|manufacturing.itemDefinition.environmentstructured:markets[],vehiclePlatforms[],productionPhase,operationalPhase,serviceModes[].component.kindenum (vehicle|system|ecu|softwareComponent|dataObject) and explicitparentComponentIdsemantics so a tree can be rendered.component.safetyRelevancewithsafetyGoalRefandasilContext(informational only — RDX does not assert ASIL ratings, only links them).assumptionobjects:{ id, statement, owner, validityPeriod, responsibleParty, evidenceRefs[], riskSharingNote }. Keep a string-array form accepted for backwards compatibility, but mark deprecated.References
spec/json/rdx.schema.json—architectureoneOf string|objectexamples/rdx-xml-comprehensive.xml— still uses string architectureAcceptance criteria
architectureandassumptionsemit deprecation warnings intools/validate.shmethodology/Methodology.mddocuments the new fields with one diagramSurfaced by external AI peer review; see chat transcript for full review text.