RSM Model
Canonical and semantic state: kinds, relationships, state categories and the canonical progression. Specified in 01–04 (01 Normative; 02–04 Draft Normative).
Developers
Applications, agents and connectors reach RSM through its protocol, never through its tables. That boundary is what lets each of them evolve without becoming a hidden system of record. This page shows where your code sits, where each part of the architecture fits, where each kind of state lives, and the rules every consumer follows.
WHERE YOUR CODE SITS (source: 05 §3.2, §25.3)
Dependencies flow toward canonical and semantic authority. Canonical domain code does not depend on a particular database, HTTP framework, vector store, graph engine or intelligence system.
The pattern to avoid is an application reading and writing RSM database tables (source: 05 §25.3). It skips validation, authority and provenance at once.
Product code, autonomous agents and connectors evolve freely. They may keep local UI state and caches.
Typed access to RSM operations. The first SDKs are Go, for servers and connectors, and TypeScript, for web applications.
The governed boundary: Actor, Principal, Subject, purpose, authority and idempotency are evaluated here.
Admits operations, keeps the operation and change journals, and maintains projections.
Authoritative RSM state. Nothing above the protocol writes to it directly.
APPLICATIONS AND AGENTS → PROTOCOL → RUNTIME → MODEL
The full path runs from applications and agents, through the RSM Protocol and its authorized operations, to the RSM Runtime and the Canonical and Semantic Model. Identity and discovery support that path from beside it. Each part is labeled with its maturity, because a specified boundary is not shipped software.
Canonical and semantic state: kinds, relationships, state categories and the canonical progression. Specified in 01–04 (01 Normative; 02–04 Draft Normative).
Authorized operations, admission, journals, projections and federation. Specified in 04 and 05 (Draft Normative). The reference implementation is planned (07), not shipped.
Identity, trust, issuance and resolution. Proposed specifications 08 and 09. How core protocol operations depend on it is proposed integration architecture.
The discovery, directory and federation network: participants discover one another, locate authoritative services and establish trusted connections across independently governed RSM Nodes. It refers and connects; it does not own what it discovers, and discovery is not authoritative resolution. Not governed by 01–09; its integration with the protocol is proposed.
Bounded, identity-bearing participants that sense, discover, reason, orchestrate and propose. They reach the Model through the protocol as DigitalAgent Actors, like any application. The agent architecture is proposed, not a governed specification.
Cognition is not authority. An agent's plan, explanation or reported completion is not evidence that an action was authorized or done. A consequential effect still needs an authorized operation, an authenticated principal acting under current authority, and an auditable result. See why capability is not permission (source: 01 §14.2).
CANONICAL, PROJECTED AND DERIVED (source: 07 §3.2–3.4)
The reference architecture separates three planes of state in code. Knowing which one you are touching tells you what you may do with it. These are runtime planes, not the three conceptual planes of the Model.
Locally authoritative state and the rules for changing it: persistence, semantic validation, identity resolution, authority, policy, revisions, provenance and the journals. No projection or intelligence component mutates it directly.
Search, graph, spatial and vector views, plus Biography over canonical history. Every projection is rebuildable, provenanced, permission-aware and non-authoritative. Deleting one never destroys canonical meaning.
Consumes authorized state and produces Derived reasoning: Configurations, Gaps, Pathways, Recommendations. It owns none of the participants' canonical state, and the reference reasoner can be replaced without changing the canonical runtime.
THE RULES EVERY CONSUMER FOLLOWS
RSM is permissive about implementation technology and strict about semantic boundaries. These rules apply to applications, agents and connectors alike; Integrators applies them at the connector.
Name similarity is not identity resolution. External identifiers map to canonical identities, and uncertain matches stay uncertain.
Every imported fact stays traceable to its source, and mappings that change meaning are versioned.
Consumers create operations, not privileged rows. Nothing becomes canonical except through the canonical mutation boundary.
A valid login is not permission to commit on behalf of a Subject. Agency answers whether an entity can act; Authority answers whether it may.
A Configuration, Gap, Pathway or Recommendation stays Derived wherever it is stored, and never stands in for a participant's Commitment.
Search, graph, spatial and vector systems are optimized views, never hidden truth. A stale projection is reconciled against its authority before anything consequential happens.
STACK BOUNDARY (source: 07 §2, §15)
The reference realization is a Go modular runtime on PostgreSQL with PostGIS, a JSON-LD, RDF and SHACL semantic contract, an HTTPS and JSON protocol described in OpenAPI 3.1, and explicit ports for search, graph, spatial, temporal, vector, embedding and model providers. It is federated externally and modular internally (source: 07 §1.3): nodes federate through the protocol, never through shared database access.
None of that defines what an RSM concept means. A technology may implement, index, project, transport or reason over part of the Model; it does not acquire authority over it. The Model shows which technology does not define which concept and the ports that keep each one replaceable. The Technology & Infrastructure Architecture (07) is an implementation architecture, not a normative specification.
FIVE PLANNED MILESTONES (source: 07 §22)
Planned in the Technology & Infrastructure Architecture. These are architectural capabilities, each cumulative and independently verifiable; they describe intended work, not shipped releases.
Reconcile the existing kernel with the normative foundation and establish the full infrastructure skeleton.
Make the canonical Model operational through the complete protocol and a rebuildable read and projection plane.
Connect independently governed nodes and existing systems without weakening authority boundaries.
Realize the derived reasoning plane while keeping intelligence and authority distinct.
Prove the complete architecture and hand it to applications and intelligence systems.
Each milestone leaves the repository fully verifiable, and every optimization must pass the same contract suites (source: 07 §24) before it replaces an adapter.
WHERE TO GO NEXT
Start with the Concept Paper and the Subject Model, and look up each term before turning it into a type. Every concept page quotes its governing definition.
Applications call RSM operations, not canonical tables. Authority and idempotency live at that boundary.
Map identity, semantics and provenance through a Connector, and keep the source system authoritative where it legitimately is.
Run the positive and negative suites. RSM is real only when invalid transitions fail at the correct boundary.