Chapter 05 · Architecture & Data Integration · Design specification

Data Sources, Semantic Alignment
& Federation

Many source systems. One governed semantic model. Preserve ownership, identity and provenance while mapping structured information into composable property graph fragments.

IFC · IDS · OWL/RDFExcel · JSONbSDD · APIsSemantic AlignmentFederated ServersControlled Sharing

Design preview 0.1 · Baseline: OpenSPAD 7.8.53 · 17 October 2026 · Federation and fine-grained access controls are future architecture, not deployed features.

05.1 / Architecture principle

Integrate by graph meaning, not by file format.

Every data source contributes to a common semantic property graph model while retaining its identity, provenance, version and access permissions. A data source is not necessarily an ontology: a project IFC or Excel file is input data; an OWL graph may be a vocabulary and relationship source.

Independent sources IFC model / schema IDS information requirements OWL / RDF ontology Excel / JSON datasets bSDD / NATSPEC REST / GraphQL / BigQuery Manufacturer / other server Canonical graph fragmentsSourceResource → EntityEntity → Property → ValueEntity → ConceptVersionProvenance → SourceVersionScope → AuthorizationMappings keep evidence and origin. Global knowledgePublished Concept revisionsOWL / IFC schema / bSDDReference Use Cases Project knowledgeIFC instances / Excel valuesUse Cases / IDS / IDM / BCFValidation runs / results
Logical target model, not a claim that all connectors, graph labels or authorization mechanisms exist in release 7.8.53.
05.2 / Connector catalog

Sources are typed. Meaning is mapped.

IFC Schema & IFC Models

Schema classes and Psets are shared semantic references; model instances, GlobalIds, quantities and properties belong to individual projects.

IDS XML

Import specifications, applicability and requirements. Map constraints to concepts; validate expressible rules and export IDS with traceability.

OWL / RDF

Classes, properties and subclass relationships. Maintain ontology URI and version; do not confuse ontology definitions with project records.

Excel / JSON

Normalize rows, columns, units, identifiers and classifications as data; import custom properties with explicit mapping assumptions.

bSDD & NATSPEC

External classification, property and dictionary catalogs. Track provider namespaces, external IDs, versions and license requirements.

REST / GraphQL

Connector-specific authentication, pagination, incremental changes, rate limits, schema evolution and transformation mappings.

Manufacturer Data

Vendors publish governed product and specification fragments without transferring full internal databases.

BigQuery / Enterprise Data

Future enterprise connection using narrow, authorized extracts or queries; never imply a live integration without proof.

Important: this table is a connector and information-model specification. It does not certify that each integration is deployed or operational.

05.3 / Canonical property graph contract

Every mapping must be explainable.

A canonical graph is an interoperability contract, not a requirement that every remote server copies the entire graph or uses identical physical storage. The target design separates references, project records and source evidence.

Identifiers

Stable internal graph IDs; source-scoped external IDs and URIs; IFC GlobalId where applicable; mapping identity and version.

Semantic references

Concept identity plus an immutable published revision; typed classes, property sets and properties; alignment confidence and reviewer decision.

Governance metadata

Provider, ingestion timestamp, source version, project/tenant scope, validity interval, access policy reference and evidence path.

Illustrative graph pattern (design, not existing schema guarantee)

(:SourceResource {uri, provider, version})
  -[:SUPPLIES]-> (:GraphFragment {fragment_id, scope})
  -[:CONTAINS]-> (:Entity {entity_id, external_id})
  -[:MAPS_TO {method, score, reviewed}]-> (:ConceptVersion {concept_id, revision})
  -[:HAS_PROPERTY]-> (:PropertyValue {value, unit, datatype})

(:UseCaseVersion)-[:REFERENCES_CONCEPT]->(:ConceptVersion)
(:ValidationResult)-[:EVALUATES]->(:Entity)

These label and edge names illustrate an architectural contract. The existing OpenSPAD Neo4j model should be mapped deliberately rather than overwritten by the diagram.

05.4 / Semantic alignment pipeline

From candidate match to governed relation.

01 · INGEST
Discover

Register connector, provider, source identity and scope.

02 · EXTRACT
Normalize

Parse classes, properties, units and external identifiers.

03 · MATCH
Propose

Rank candidate Concepts with URI, lexical, structural and vector evidence.

04 · REVIEW
Decide

Confirm, reject, defer or request a Concept correction.

05 · PERSIST
Reference

Store mapping, reviewer, confidence, source and Concept revision.

06 · MONITOR
Reassess

Detect upstream changes and enqueue impact analysis.

Worked example: external door classification

Source class
Provider: bSDD
External Door
→
Reviewed alignment
URI / class / semantic evidence
Maps to IFC Door
→
Project view
IFC Door → Psets → Properties

The alignment does not convert every bSDD item into a new root concept. Review and typed rules determine placement. Confidence scores are proposals, not permissions.

Governance rule: creating or correcting shared global Concepts requires Administrator review. A BIM Manager may propose changes for project needs but cannot silently change published global versions.
05.5 / Scope boundaries

Global knowledge ≠ project facts.

Global / reusableProject-specificBoundary rule
OWL ontologies and IFC schema definitionsImported IFC building elements and their property valuesReference ontology/schema revisions; do not promote project facts automatically.
Reviewed bSDD classifications and published ConceptsExcel imports, custom Psets, local alignmentsKeep project-scoped mapping and property values attached to the project.
Reference Use Cases and published graph rulesProject Use Cases, IDS, IDM, BPMN, BCF topicsProject compositions pin their chosen shared revisions.
Published shared transformationsValidation runs, snapshots, findings and viewer linksResults remain traceable to source versions and model revisions.
05.6 / Distributed architecture

A logical graph across independently controlled servers.

Federation means authorized data and semantic references can be resolved across servers. It does not require a single physical central Neo4j database, blind replication or permission-free traversal.

OpenSPAD Server AGlobal Concept libraryPublished revisionsOpenSPAD Server BManufacturer catalogLimited shared subsetOpenSPAD Server CProject-controlled IFCIDS / Use Cases / BCFPolicy-aware logical federationIdentity · discovery · query planningAuthorized traversal · provenance · version pinningComposed project information
Target architecture / FUTURE. Multi-server distributed query, authorization and selective publication are not claimed as released functionality.

Remote reference

Use stable server/resource identifiers plus revision selectors. Resolve remote nodes only through authorized interfaces.

Controlled replication

Optional policy-driven materialized snapshots with provenance and expiry; no default full graph copying.

Failure behavior

Surface offline providers, partial graph completeness, stale mappings and unavailable permissions explicitly.

05.7 / Information requirements

IDS is both a source and a deliverable.

INPUT
Import IDS

Read XML specification, applicability, facets and requirements.

NORMALIZE
Link semantics

Attach constraints to Concept revisions and IFC entities where expressible.

COMPOSE
Project Use Case

Edit requirements in a governed project information profile.

OUTPUT
Export IDS

Write schema-valid XML for supported constraints with provenance.

VERIFY
Validation run

Record PASS/FAIL/NOT_EVALUATED and IFC GlobalId references.

A lossless IDS round trip and cross-object/geometric rule checks are goals, not capabilities proven by the current pilot. Keep unsupported constraints explicit instead of silently omitting them.

05.8 / Source change and release governance

A new source must not silently mutate old Use Cases.

01
Change detected

New ontology or dictionary version, changed API data or additional manufacturer product.

02
Difference captured

Compare source identities, semantic claims, classes and properties.

03
Impact proposal

Identify affected mappings, Concepts, rules and dependent Use Case versions.

04
Admin decision

Reject, defer, map to existing revision, or create a new draft revision.

05
Publish & migrate

Published Concept versions remain immutable; existing Use Cases upgrade only by explicit choice and revalidation.

Recommended impact report fields

source resource + versionadded / changed / removedcandidate Conceptsaffected Use Case revisionsconfidence and human reviewproposed actionvalidation consequencesaudit trail
05.9 / Access governance

Visibility follows the resource, not just the link.

Future multi-tenant federation must evaluate organization, provider, server, project, object, revision and action-level policies. A visible cross-server relationship must never imply access to the remote node or its properties.

Administrator

Manage globally shared source registrations, concept publishing and alignment governance within assigned authority.

BIM Manager

Import project data, compose project information requirements, submit mapping or Concept change proposals.

Standard User

Compose or reference approved Use Cases within permitted projects, without modifying the shared Concept library.

Provider / Manufacturer

Choose the subset of product data shared to authorized partners and revoke or update publication deliberately.

Design review checklist: protect against data leaks through graph traversal, search rankings, embeddings, previews, caches, exports and diagnostic logs. Enforcement is not yet a verified deployed feature.

05.10 / Evidence and maturity

Document what runs; label what is intended.

ComponentStatusEvidence / limitation
Neo4j graph runtime and persisted Concept rule pilotVerified baseline / PilotSeparate productive Neo4j installation and persisted ALLOWS_CHILD pilot in 7.8.53; not general policy enforcement.
Concept library, source inspection and OWL / IFC / IDS explorationsPartial / PilotHistorical module experiments and ontology/mapping workflows; individual deployment and end-to-end behavior must be rechecked.
Unified ingestion and reviewed semantic alignment workflowPlannedTarget data contract; no claim that all listed connectors are production-ready.
IDS lossless round trip, model checks and 3D issue highlightingPlannedRequires supported-facet coverage, IFC validation implementation and evidence.
Federated cross-server query and selective publicationFutureArchitecture vision only; remote permission and synchronization enforcement pending.
Multi-tenancy and cross-server securityFutureThreat modeling, policy model, test harness and independent verification required.

Source of maturity assertions: OpenSPAD development snapshot through 7.8.53 and agreed architecture decisions as of 17 October 2026. This document is a design specification, not a verified installation inventory.