The context & data layer for TACITUS

The trust graph for conflict intelligence

DIALECTICA is the context, foundation, and data layer for TACITUS. It structures any conflict — workplace friction, commercial disputes, geopolitical crises — into a deterministic knowledge graph grounded in 15 peer-reviewed frameworks. Other apps build on top. DIALECTICA is the foundation they reason over.

Layer 1

Context Layer

We extract actors, interests, power dynamics, emotions, trust states, and causal chains from unstructured text.

Layer 2

Foundation Layer

25+ deterministic symbolic rules fire against the Conflict Grammar ontology (15 node types, 20 edge types) — treaty violations, escalation thresholds, ripeness conditions are computed, never hallucinated.

Layer 3

Reasoning Layer

6 AI agents reason over structured graphs, not raw text. Neural predictions are firewalled — deterministic conclusions are never overridden by probabilistic ones.

TACITUS Ecosystem

DIALECTICA is the trust graph

DIALECTICA is not a standalone tool — it's the context and data layer for the entire TACITUS platform. It extracts, structures, and validates conflict data into a deterministic knowledge graph. Other applications in the TACITUS ecosystem build on top of this foundation to deliver specialized capabilities.

Think of it like this: DIALECTICA is to conflict intelligence what a database is to an application. It provides the structured, queryable, deterministic truth that every upstream tool — mediation assistants, risk dashboards, policy simulators, negotiation trainers — can reason over with confidence.

Architecture

Apps

Mediation tools, risk dashboards, negotiation aids, policy simulators

DIALECTICA

Trust graph · Context layer · Deterministic foundation

Data Sources

Documents, transcripts, reports, news, HR complaints, legal filings

Why LLMs need a deterministic foundation for conflict

Language models guess. DIALECTICA computes. Here's the difference.

QuestionLLM AloneLLM + DIALECTICA
"What stage is this conflict at?"Guesses based on keywordsComputes Glasl stage from event chains + escalation velocity
"Are there treaty violations?"Summarizes text about violationsDeterministic pattern matching against norm graph
"Who has leverage?"Describes who seems powerfulFrench & Raven power analysis with magnitude scores
"Is this conflict ripe for resolution?"General assessmentZartman ripeness model: MHS + WO computed from graph
"What are the underlying interests?"Extracts stated positionsMaps stated vs unstated interests with BATNA analysis
"How might this escalate?"Speculative narrativeCausal chain analysis with mechanism tagging (escalation, retaliation, contagion)

Grounded in peer-reviewed conflict research

Every node type, edge type, and symbolic rule in DIALECTICA traces to published academic work.

Glasl, F. (1982). The Process of Conflict Escalation and Roles of Third Parties.”

9-stage model, basis for escalation detection

Fisher, R. & Ury, W. (1981). Getting to Yes.”

Interest-based negotiation, BATNA framework

Zartman, I.W. (2000). Ripeness: The Hurting Stalemate and Beyond.”

Ripeness theory for intervention timing

Galtung, J. (1969). Violence, Peace, and Peace Research.”

Direct/structural/cultural violence triangle

French, J. & Raven, B. (1959). The Bases of Social Power.”

8 power domains

Mayer, R., Davis, J., & Schoorman, F. (1995). Model of Organizational Trust.”

Trust = f(ability, benevolence, integrity)

Plutchik, R. (1980). A General Psychoevolutionary Theory of Emotion.”

8 primary emotions wheel

Pearl, J. (2009). Causality.”

Causal inference for event chain analysis

Beyond RAG

Why ontology-augmented generation outperforms RAGfor structured domains

RAG retrieves chunks of text. OAG retrieves structured knowledge with typed relationships and computed properties.

In conflict domains, relationships between entities ARE the data. A vector search can find mentions of ‘Iran’ and ‘sanctions’ — but it cannot compute that sanctions CAUSED enrichment breach which CAUSED escalation to Glasl stage 6.

Deterministic symbolic rules provide guarantees that probabilistic models cannot: if a treaty exists and an event violates its terms, that violation is a fact, not a prediction.

Academic backing

Pan et al. (2024). “Unifying Large Language Models and Knowledge Graphs: A Roadmap.”

Argues for structured knowledge integration over retrieval-only approaches. LLMs + KGs = complementary strengths.

Ji et al. (2022). “Survey on Knowledge Graphs: Representation, Acquisition, and Applications.”

Ontology-driven knowledge representation outperforms unstructured retrieval for domains with rich relational semantics.

RAG

  • Retrieves text chunks
  • No relationship types
  • Cannot compute causality

OAG

  • Typed knowledge graph
  • 20 edge types, computed
  • Deterministic reasoning

Architecture Preview

A neurosymbolic pipeline from raw text to decision support in seconds.

Unstructured Text

Input

GLiNER NER

Pre-filter

Gemini Extraction

Structured

Pydantic Validation

Type-safe

Neo4j Knowledge Graph

Graph DB

Symbolic Rules

Deterministic

AI Agents

6 specialists

Decision Support

Output

Built for Developers

Install the ontology, hit the API, or connect via MCP.

Python

pip install dialectica-ontology

Node / TypeScript

npm install @tacitus/dialectica-sdk

Claude Desktop (MCP)

MCP Server for Claude Desktop