TACITUS is conflict intelligence infrastructure — a governed data layer that transforms scattered dispute traces into structured knowledge graphs with temporality, causality, and provenance preserved.

How is TACITUS different from standard LLMs?

Standard LLMs lack persistent memory, cannot track temporal ordering of events, and hallucinate on conflict facts. TACITUS provides a deterministic knowledge graph layer that grounds LLM reasoning.

What industries does TACITUS serve?

TACITUS serves HR disputes, legal litigation, commercial conflicts, diplomatic negotiations, institutional governance, and humanitarian conflict analysis.

What products are in the TACITUS suite?

The TACITUS suite includes four products: PRAXIS (the full conflict operating system), Conflict Compass (accessible conflict navigation), Windtunnel (scenario simulation), and Centaurus (deep conflict intelligence for professional mediators). All are experimental demos built on the same Core Engine.

The Conflict Graph EngineIn Active Development

Temporality. Causality. Provenance.

The Core Engine transforms scattered conflict traces into a unified graph structure that preserves what matters — and exposes it to deterministic reasoning.

Foundation

The Three Pillars

Temporality

Every event is a node with a timestamp. Every claim is version-stamped. The graph can reconstruct exactly what was said, when, and in what sequence — months or years later.

Event sequencing with DAG preservation
Claim evolution tracking over time
Position shift detection
Causal timeline reconstruction

Causality

The graph stores not just 'A happened then B' but 'A caused B via mechanism M, involving actors X and Y.' That distinction is the difference between a log and intelligence.

Causal chain encoding
Escalation pathway mapping
Trigger identification
Counterfactual pathway modeling

Provenance

Every assertion in the graph traces back to a source document, timestamp, and actor. Nothing is asserted without evidence. The graph is a full audit trail by design.

Source attribution on every node
Evidence linking to claims
Compliance-ready audit trails
Chain-of-custody for conflict data

Ontology

The ACO: Mediator-Grade Reasoning Structure

The Agentic Conflict Ontology encodes the primitives that human conflict professionals actually use — making AI reasoning structurally rigorous.

Actor Classification

PrincipalPrimary party with direct stake

AgentActing on behalf of a Principal

VetoPlayerCan block resolution

BridgeActorConnects otherwise disconnected parties

MediatorNeutral facilitation role

Claim-Evidence Binding

PositionWhat the actor says they want

InterestWhat they actually need

RedLineNon-negotiable boundary

BATNABest alternative to negotiated agreement

EvidenceSupporting material for a claim

Causal Relation Types

TriggeredByDirect causal trigger

EscalatedFromDerived escalation path

BlockedByActive inhibitor

EnabledByPrerequisite or facilitator

conflict-ontology.ts

// Agentic Conflict Ontology — TypeScript enums

enum ActorRole {
  Principal   = "PRINCIPAL",
  Agent       = "AGENT",
  VetoPlayer  = "VETO_PLAYER",
  BridgeActor = "BRIDGE_ACTOR",
  Mediator    = "MEDIATOR",
}

enum ClaimType {
  Position = "POSITION",
  Interest = "INTEREST",
  RedLine  = "RED_LINE",
  BATNA    = "BATNA",
  Evidence = "EVIDENCE",
}

enum CausalRelation {
  TriggeredBy   = "TRIGGERED_BY",
  EscalatedFrom = "ESCALATED_FROM",
  BlockedBy     = "BLOCKED_BY",
  EnabledBy     = "ENABLED_BY",
}

interface GraphNode {
  id: string;
  type: "Actor" | "Claim" | "Event" | "Evidence";
  role?: ActorRole;
  claimType?: ClaimType;
  timestamp?: string;          // ISO 8601
  sourceId: string;            // provenance
  confidence: number;          // 0–1
}

Graph Architecture

Where AI Reasoning Gets Grounded

Nodes

What Exists

Actors
Claims
Events
Evidence
Constraints
Commitments

Edges

How They Connect

ASSERTS
TRIGGERED_BY
ESCALATED_FROM
PROVIDES_EVIDENCE
VIOLATED
ENABLED_BY

Layers

Different Lenses

Temporal layer (GND)
Actor network (CTX)
Evidence layer (EVD)
Reasoning layer (RZN)

Comparison

Graphs vs. Vectors: Why Architecture Matters

Vector Embeddings — What's Lost

✗ Causality destroyed
✗ Temporality flattened
✗ Provenance impossible
✗ Contradictions invisible
✗ Time-sensitive facts merged
✗ No audit trail

Conflict Graphs — What's Preserved

✓ Causal chains intact
✓ Temporal ordering preserved
✓ Full audit trails
✓ Explicit contradictions
✓ Version-stamped claims
✓ Source attribution on every node

Pipeline

From Scattered Traces to Structured Memory

01SourcesEmails, transcripts, docs, notes

02ParseText segmentation & entity detection

03ACO MappingConflict ontology type binding

04Graph ObjectsNode & edge creation

05Causal LinkingRelation & temporal ordering

06OutputAPI-queryable, LLM-groundable graph

Design Philosophy

Reasoning Like the Professionals

The ACO encodes the reasoning patterns of the world's most experienced conflict professionals — not generic NLP categories.

Mediators

Interest vs. position separation
BATNA surface & comparison
Zone of Possible Agreement mapping
Bridging actor identification

Conflict Analysts

Escalation pathway encoding
Actor network & alignment mapping
Trigger event identification
Cross-case pattern recognition

Diplomats

Position evolution tracking
Coalition formation analysis
RedLine & commitment mapping
Narrative divergence detection

Ready to Build on the Core Engine?

Request a pilot demo or explore the developer documentation.

Request a Pilot Demo View API Documentation

◳

// Agentic Conflict Ontology — TypeScript enums enum ActorRole { Principal = "PRINCIPAL", Agent = "AGENT", VetoPlayer = "VETO_PLAYER", BridgeActor = "BRIDGE_ACTOR", Mediator = "MEDIATOR", } enum ClaimType { Position = "POSITION", Interest = "INTEREST", RedLine = "RED_LINE", BATNA = "BATNA", Evidence = "EVIDENCE", } enum CausalRelation { TriggeredBy = "TRIGGERED_BY", EscalatedFrom = "ESCALATED_FROM", BlockedBy = "BLOCKED_BY", EnabledBy = "ENABLED_BY", } interface GraphNode { id: string; type: "Actor" | "Claim" | "Event" | "Evidence"; role?: ActorRole; claimType?: ClaimType; timestamp?: string; // ISO 8601 sourceId: string; // provenance confidence: number; // 0–1 }