Source Graph

Source Graph records source scopes, graph versions, reviewed relationships, and citation anchors for regulated source material. It is a generic grounding foundation for policies, standards, contracts, procedures, datasets, evidence packs, and other governed corpora.

Use it when teams need source-grounded comparison, cross-reference mapping, citation readiness, and downstream evidence for document treatment or audit packages.

What Source Graph Owns

Source graph scopes that describe a bounded source set.
Versioned graph build and readiness posture.
Source, citation, rights, and evidence drift posture for graph versions.
Graph nodes linked to source-owner refs or neutral concepts.
Reviewed graph edges such as references, supersedes, amends, conflicts with, supports, maps to, depends on, duplicates, equivalent term, and implementation link.
Citation anchors pointing to source-owner refs, extracted chunks, page or region anchors, or future source-owner references, with review status before downstream use.

Source Graph stores refs, status, provenance, confidence, rights, and limitation snapshots. It does not store raw source text, OCR text, prompts, provider payloads, vectors, storage keys, connector secrets, or protected standard/body text.

User Journey

Create a source graph scope for the source set or corpus being reviewed.
Record a graph version with source, build, readiness, and optional AI-assistance posture.
Record nodes for source items, sections, clauses, requirements, datasets, indicators, or neutral concepts.
Record proposed relationships between nodes.
Review proposed relationships with a governed reason before they are treated as accepted graph evidence.
Record citation anchors that point back to source-owner or extraction refs.
Activate, block, or retire citation anchors with a governed reason before downstream use.
Assess graph-version drift when source-owner refs, relationships, or citations change. The assessment updates version readiness with states such as current, source-newer-version, source-withdrawn, relationship-review-required, citation-review-required, relationship-stale, citation-stale, rights-blocked, or evidence-required.

Manual graph and citation evidence works in no-AI deployments. AI-assisted candidate generation, semantic retrieval, connector ingestion, and public APIs are separate capabilities and must show no-AI profile, provider-not-configured, deployment-dependent, roadmap, or another accurate capability state when unavailable.

Boundaries

Source Graph references other owners instead of replacing them:

Governed Source owns source authority, item, version, rights, trust, and lifecycle truth.
DMS owns document, version, folder, artifact, and access truth.
Content Extraction owns extraction, chunks, search, preview, OCR posture, and AI handoff readiness.
AI owns runs, suggestions, retrieval, embeddings, vectors, provider/profile policy, and optional candidate generation.
Governance Trace owns neutral cross-module trace links for runtime governed records.
Audit Packages, Work Management, and Compliance Posture consume Source Graph descriptors; Source Graph does not create local export, task, or posture engines.

Safe Evidence

Use Source Graph evidence for:

relationship review status;
citation review status;
graph version readiness;
graph version drift status;
citation coverage posture;
source and extraction refs;
confidence and limitation snapshots;
audit package and readiness evidence.

Do not use it to claim legal validity, official interpretation, global citation accuracy, certified comparison, confidential vector search, or public integration API availability unless a separate implemented capability provides that evidence.

Drift assessment uses safe source refs, relationship statuses, and citation refs. It does not copy source text, OCR text, prompts, provider payloads, vectors, embeddings, storage keys, or connector secrets into graph metadata.

Cloud And On-Prem

Source Graph is available through the same workspace behavior in cloud and on-prem deployments. Deployment differences should appear as capability states or provider configuration evidence, not as different product semantics.