A neurosymbolic engine for the people who manage disagreement , for a living, or by accident.
Conflict is an information-asymmetry problem. TACITUS is the infrastructure that closes the gap.
Human-friction professionals need better responses. HR leaders, mediators, legal counsel, policy analysts, peace-process teams , generic AI only goes so far before the hallucinations cost more than they save. TACITUS is the infrastructure designed for the work they actually do.
8 primitives · 41+ ontology classes · 4 graph layers · 5 products · Open spec · Under active development
Alex publicly criticised Maya on an internal channel, asserting that Maya missed the Q2 deadline. Maya escalated to HR, filing a complaint on May 14. Alex responded that continued collaboration depends on a written apology.
{ "actors": 0, "claims": 0, "commitments": 0, "events": 0 }
FIND YOUR ENTRY POINT
Four concrete groups of users. One engine underneath. Pick the row that matches the work you actually do , each card lists the roles we are building for and the first product to try.
WHY CONFLICT · WHY NOW
Parties in a dispute are not usually irrational. They are operating on different pictures of the same situation , different timelines, different causal stories, different readings of who committed to what. Strip the asymmetry away and most arguments dissolve; what remains is a hard decision, not a deadlock. TACITUS is the infrastructure that surfaces the structure underneath the disagreement, so the hard decision is the thing humans actually work on.
HR leaders
Grievances, escalations, team disputes
Mediators & ombuds
Multi-party sessions, commitment tracking
Risk & compliance officers
Whistleblower reviews, governance breaches
Legal & ADR counsel
Pre-litigation mapping, ADR strategy
Policy analysts
Stakeholder modelling, policy reception
Research & intelligence analysts
Archive-scale document synthesis
Communications & crisis leads
Message stress-testing, reframing detection
Peacebuilding & humanitarian teams
Multi-actor case files, evolving commitments
“Finally, a tool that treats a dispute like the structured object it is.”
THE PROBLEM
The failures are not bugs. They are architectural properties.
Temporality
Transformers see sequences, not timelines. Ask when a commitment was made, broken, or re-negotiated and a generic LLM guesses plausible dates. Dialectica maintains a temporal DAG under the graph.
Causality
Statistical co-occurrence is not causation. Ask which event caused which response, across three actors and six turns, and a generic LLM invents connections. Dialectica stores causal edges as first-class graph objects.
Provenance
A fluent paragraph with no citation is an assertion, not evidence. Dialectica binds every extracted primitive to its source span, so every downstream claim is auditable back to the original document.
The Keshara Valley Ceasefire Crisis is not a real dispute. It is a worked example , constructed to show every primitive, every edge type, and at least one load-bearing contradiction.
CASE BRIEFING · FICTIONAL
Three parties, four weeks, one collapsed agreement. The Government of Taruna and the Keshara Liberation Front (KLF) signed a ceasefire in January, mediated by UNSOM with local Elder Council witnessing. Three weeks later the village of Mira was raided; each side blamed the other; the emergency session failed to reconcile contradictory accounts; by Feb 08 the ceasefire had formally collapsed.
Below is the same crisis as a TACITUS graph. Hover a node to see its edges. Click to pin. The primitives below are color-coded; the timeline along the top is the temporal DAG underneath.
EVENT TIMELINE · TEMPORAL DAG
Ceasefire signed in Nairobi by KLF, Gov. of Taruna, under UNSOM mediation
Village of Mira raided; KLF and Gov. make contradictory claims about initiator
Emergency guarantor session convened; mediators surface both claims with source provenance
Ceasefire collapses. Commitment cm1 (weapons withdrawal) formally violated
WHAT THE GRAPH SURFACES
CONTRADICTION
Gov. asserts KLF violated the ceasefire on Feb 03. KLF asserts Gov. entered the DMZ first the same day. Same event, same date, opposite attributions. The engine flags this as a load-bearing contradiction; any generation downstream is forced to cite both.
INTEREST-POSITION GAP
Surface positions read as an armed-group dispute. Primitive-separated, the underlying interest is regional autonomy , which opens resolution pathways that never appear at the position layer.
COMMITMENT DRIFT
Even as the ceasefire collapses, commitment cm2 (humanitarian corridor) remains active per the signed annex §4. A mediator without this structure would read the situation as total collapse.
TEMPORAL CHAIN
Edges trigger, escalate_to, and violates connect four events in five days. Not just correlation , each edge is typed, and each carries the source it was extracted from.
HEAD-TO-HEAD · ASK A QUESTION ABOUT THIS CASE
Three questions a mediator or analyst would actually ask about Keshara Valley. Standard LLMs output a confident paragraph. TACITUS answers from the graph, with typed nodes and source citations.
Who violated which commitment, and when?
Generic LLM
The KLF and the Government accuse each other of breaking the ceasefire around early February. The exact sequence is unclear from the text.
No commitment ID, no edge type, no timestamp.
TACITUS (graph-grounded)
Commitment cm1 (weapons withdrawal, 30-day deadline 2024-02-14, ceasefire annex §2.1) was marked VIOLATED on 2024-02-03 by event e2 (village of Mira raid). Edge type: VIOLATES. Responsibility is contested between actors a2 (KLF) and a3 (Gov. of Taruna) via contradictory claims c1 and c2.
Is the humanitarian corridor still active?
Generic LLM
The ceasefire seems to have collapsed around Feb 8, so most agreements are likely null.
Conflates two distinct commitments with different statuses.
TACITUS (graph-grounded)
Commitment cm2 (humanitarian corridor, ceasefire annex §4) remains status=ACTIVE despite cm1 being violated. Distinct commitment nodes; no cascading invalidation rule applies. The Elder Council (a4) continues to endorse (c3) per annex §4.2.
What interest underlies the KLF position, and where does resolution space live?
Generic LLM
The KLF appears to want autonomy, but the details are beyond the provided text.
Cannot separate Position from Interest at the type level.
TACITUS (graph-grounded)
KLF (a2) holds Interest i1 = "Autonomy for the Valley" (source: manifesto + interviews). This is distinct from their Position on the ceasefire. Gov. (a3) holds Interest i2 = "Territorial integrity" (source: constitutional ref. Art. 1). Resolution space sits in the overlap: federated autonomy arrangements that preserve territorial integrity remain unexplored in the graph and would be the obvious mediator focus.
WHY THIS HELPS LLMS
STEP 1
Ask 'who violated what, and when?' , the graph returns typed nodes and edges with timestamps and source-document spans. No statistical guess; the answer is a database read.
STEP 2
Downstream language models receive ontology-shaped context , Actor X asserted Claim Y against Commitment Z , not a wall of prose. Fewer tokens wasted reconstructing structure the graph already knows.
STEP 3
Every extracted primitive carries provenance back to the source span. Any LLM output derived from the graph is natively citable , hallucination on keyed facts is architecturally harder.
STANDARD LLM FLOW
Input prose → attention over tokens → fluent answer without verifiable ground
OAG FLOW (TACITUS)
Input prose → extract → typed graph → deterministic query → fluent answer with source-bound citations
This is what the term neurosymbolic means in practice. The graph carries the structural weight; the language model stays in charge of the language. Each covers the other’s blind spot.
41+ classes · 29+ properties · Pydantic + OWL/Turtle · pip install tacitus-ontology
Explore the full ontology →ONTOLOGY PLAYGROUND · INTERACTIVE
Pick one of three canonical samples, click Structure it, and watch Dialectica tag every primitive , actors, claims, interests, commitments, constraints, leverage, events, narratives , with full provenance.
INPUT · dispute text
Public criticism on an internal channel, escalated to HR.
Alex publicly criticised Maya on an internal channel on May 12, asserting that Maya missed the Q2 deadline. Maya escalated to HR on May 14, filing a complaint that Alex violated the professionalism policy. Maya wants to keep her role and a written apology. Alex has said continued collaboration depends on a formal written apology. HR is bound by the grievance procedure §4.
8 PRIMITIVES
GRAPH STATS · typed output
Every node is bound to the source span it came from. Every edge is typed. Every claim is auditable back to the input.
HOW IT WORKS
Every TACITUS product runs on the same three-step flow. The difference is only the surface you touch.
01 · INGEST
Documents, transcripts, emails, case files, sensor feeds. Whatever holds the trace of a dispute. TACITUS parses, segments, and anchors every span to its source.
02 · STRUCTURE
Dialectica extracts the eight primitives and writes them into a typed knowledge graph. Temporal DAG, causal edges, provenance bindings , all preserved, all queryable.
03 · REASON
Every product queries the same graph. Language models generate fluent summaries grounded in deterministic structure. Claims trace back to the original source span.
One data layer. Every product reads from it. Every query traces to source.
THE PRODUCT SUITE
Every TACITUS product reads and writes the same typed conflict graph. Structure built in one surface is instantly available in all the others. Pick the surface that matches the work you actually do.
All products are under active development. Experimental. Send feedback.
Five capabilities that only emerge when the graph is part of the reasoning.
| Capability | Standard LLM | RAG | TACITUS OAG ◳ |
|---|---|---|---|
| Multi-hop reasoning | Single-pass inference only | Retrieval window limits scope | Typed relation traversal across actors, claims, events |
| Persistent conflict memory | No memory between sessions | Chunk-level recall only | Full graph-native persistent memory |
| Temporal event ordering | Time is flattened | Only if metadata preserved | DAG-preserved ordering with version-stamped claims |
| Source provenance tracking | None | Chunk-level attribution | Node-level: assertion → source, timestamp, extractor |
| Hallucination prevention on keyed facts | Prone to hallucination | Reduced but not eliminated | Graph-grounded , deterministic on keyed facts |
Standard LLM = blind. RAG = nearsighted. TACITUS = grounded.
OAG is in active development. Open-source pipeline.
WHERE TO GO NEXT