Skip to main content

Knowledge Graph

Adaptive Brain Visualization and Semantic Intelligence Fusion — Phase 41

Purpose​

The Knowledge Graph capability provides BASTION's semantic intelligence substrate — a living graph of entities, relationships, and confidence scores extracted from planning documents and intelligence feeds. Visualized through the Brain Canvas on the Understand tab, it transforms raw text into a force-directed neural graph where analysts can explore connections, assess confidence, and discover intelligence gaps.

The brain is not a static database — it evolves as new documents are ingested, entities are resolved across sources, and confidence scores shift with new evidence.

Components​

Adaptive Brain Visualization​

The Brain Canvas renders the knowledge graph as a force-directed neural layout:

  • Node types distinguished by shape:
    • Circles: Actors (persons, organizations, military units)
    • Diamonds: Objectives and goals
    • Squares: Documents and sources
    • Hexagons: Concepts and capabilities
  • Node colors by actor category:
    • Blue: Friendly forces
    • Red: Adversary forces
    • Gray: Neutral entities
    • Green: Partner forces
    • Gold: Objectives
  • Node size proportional to connectivity degree (importance).
  • Confidence glow: Bright glow on high-confidence entities, dim on low-confidence, dashed outline for intelligence gaps (unverified entities).
  • Edge thickness proportional to relationship confidence weight.

BrainToolbar​

Controls along the canvas:

  • Clustering mode toggle: Container (group by category), Force (physics-based), Timeline (temporal arrangement).
  • Search bar: Full-text entity search with instant highlight.
  • Timeline scrubber: Slide through time to see the graph at different points — useful for tracking how the intelligence picture evolved across exercise phases.
  • Zoom and pan controls.

JSON-LD Semantic Triples​

Entities are stored as JSON-LD linked data:

  • Valid JSON (parseable by any application) and valid RDF (compatible with semantic reasoning tools).
  • Triples follow subject-predicate-object pattern: {PLA Southern Theater Command} → {commands} → {Eastern Theater Naval Forces}
  • Typed using military planning vocabulary: units, capabilities, objectives, lines of effort, tasks, constraints.
  • Compatible with W3C standards for cross-system intelligence sharing.

Entity Resolution​

When multiple documents reference the same real-world entity using different names, abbreviations, or transliterations:

  • String similarity matching identifies potential duplicates.
  • Context embedding similarity compares surrounding text.
  • Graph structural evidence (shared neighbors) supports merge decisions.
  • Analysts review merge proposals and confirm or reject.
  • Merged entities retain all source provenance for audit.

Confidence Scoring with NATO Admiralty Code​

Every entity carries a confidence assessment based on the NATO Admiralty Code:

  • Source Reliability (A through F): How reliable is the source?
    • A: Completely Reliable, B: Usually Reliable, ..., F: Cannot Be Judged
  • Information Quality (1 through 6): How accurate is the information?
    • 1: Confirmed, 2: Probably True, ..., 6: Cannot Be Judged
  • Combined rating (e.g., "B-2: Usually Reliable / Probably True") displayed on each entity's detail card.
  • Conflicting claims from different sources produce competing entity attributes with differential confidence — analysts see the full evidentiary picture.

Brain Timeline​

  • Temporal reasoning: entities and relationships are timestamped.
  • The timeline scrubber lets analysts see the graph state at any point.
  • Useful for tracking how adversary posture evolved across exercise phases.
  • Supports "what changed since last assessment" queries.

Subspace Concept​

For large graphs with hundreds of entities:

  • Subspaces are analyst-defined focus views that filter the graph to a subset of entity types, categories, or geographic regions.
  • Subspaces reduce cognitive load without losing the full graph.
  • Multiple subspaces can be saved and recalled.

Document Intelligence Pipeline​

The brain is fed by an autonomous 10-specialist agent team that processes uploaded documents:

  1. Scoping Interview: AI determines document type and extraction strategy.
  2. ExtractionTheater: Live pipeline visualization showing agent activity with SSE particle animation flowing from sidebar into brain canvas.
  3. Entity extraction: Named entities, relationships, and attributes pulled from text.
  4. OSINT gap filling: Agents query external sources to fill identified intelligence gaps.
  5. Graph merge: New entities integrated with existing graph, entity resolution applied.

Role Access​

RoleAccess
CommanderViews brain for strategic assessment. Receives intelligence summaries.
J2 IntelligencePrimary user. Manages entity resolution, confidence updates, and gap analysis.
J3 OperationsViews brain for operational planning context. Links entities to plans.
AnalystProcesses documents through pipeline. Reviews entity extractions.

Data Flow​

Document Upload (Understand Tab)
        |
        v
  Scoping Interview (AI determines strategy)
        |
        v
  ExtractionTheater (10-agent pipeline)
        |
        v
  Entity Extraction + JSON-LD Serialization
        |
        v
  Entity Resolution + Confidence Scoring
        |
        v
  +-----------------------------+
  | Brain Canvas (Neural Graph) |
  | Force-directed layout       |
  | Confidence glow + shapes    |
  | Timeline scrubber           |
  +-----------------------------+
        |
        v
  Plan Tab (entity references)
  COP Tab (entity-symbol linkage)
  Assess Tab (intelligence readiness)

Doctrinal Reference​

  • JP 2-0, Joint Intelligence — All-source analysis and fusion
  • ATP 2-33.4, Intelligence Analysis — Structured analytic techniques
  • BASTION Phase 41: Knowledge Graph and Brain Visualization
  • See also: Understand Tab — Brain is the centerpiece of this tab

Part of the BASTION Capability Tabs documentation.