Loading TACITUS
Loading TACITUS
ENGINE · ARCHITECTURE
A seven-graph neurosymbolic architecture that implements Ontology-Augmented Generation against a kernel ontology with dynamic extensions. The structure outside the model.
DIALECTICA is the engine specification underneath every TACITUS product. PRAXIS reads its Analytical and Commitment graphs; CONCORDIA writes Case-scoped subgraphs during live deliberation; Wind Tunnel forks Scenario subgraphs for reception modelling; ARGUS writes Base and Evidence on ingest. The boundaries between products are surface-level; the engine is the same.
NEUROSYMBOLIC · OPTION 2
Garcez and Lamb characterised neurosymbolic AI as the third wave, with several integration patterns. DIALECTICA is an Option-2 architecture[]: a neural system (the LLM, doing pattern recognition over noisy text) interacting with a separate symbolic reasoning system (the typed graph plus its validator). The two components cooperate by construction rather than by patching.
The neural side performs the task LLMs are reliable at: extracting structure from unstructured text. The symbolic side performs the task LLMs are not reliable at: maintaining typed structure that does not drift under self-generated context[, ]. This is what makes the engine inheritable across analysts and sessions, and what lets a partner institution verify that any Analytical claim traces back to an immutable source span.
SEVEN GRAPHS
The partition is the engineering thesis. Each graph has a different mutability contract because each serves a different query workload — and because conflating them is how every adjacent system loses its audit trail.
| Graph | Content | Update discipline | Mutability |
|---|---|---|---|
| Base | Raw ingested facts as extracted from source corpora. | Append-only at ingest. | Immutable |
| Evidence | Source spans, citations, dataset references. | Immutable on write; cryptographic-style binding to Base. | Immutable |
| Analytical | Typed primitive instances and edges — the working ACO subgraph. | Bi-temporal; revisions logged as new edges with new t_tx; invalidation, not deletion. | Append-only at t_tx |
| Narrative | Framings, drift chains, competing stories. | Drift-tolerant by design; all framings preserved. | Append-only |
| Commitment | Bi-temporal commitment register with status transitions. | StatusTransition records; revision chains preserved. | Append-only at t_tx |
| Case | Practitioner-facing subgraph for a particular dossier. | Forkable per case; case-scoped extensions. | Mutable within case |
| Scenario | Counterfactual or prospective subgraphs for options analysis. | Forked from Case; isolated; never merged back without explicit promotion. | Mutable within scenario |
Base
Raw ingested facts as extracted from source corpora.
Append-only at ingest.
Immutable
Evidence
Source spans, citations, dataset references.
Immutable on write; cryptographic-style binding to Base.
Immutable
Analytical
Typed primitive instances and edges — the working ACO subgraph.
Bi-temporal; revisions logged as new edges with new t_tx; invalidation, not deletion.
Append-only at t_tx
Narrative
Framings, drift chains, competing stories.
Drift-tolerant by design; all framings preserved.
Append-only
Commitment
Bi-temporal commitment register with status transitions.
StatusTransition records; revision chains preserved.
Append-only at t_tx
Case
Practitioner-facing subgraph for a particular dossier.
Forkable per case; case-scoped extensions.
Mutable within case
Scenario
Counterfactual or prospective subgraphs for options analysis.
Forked from Case; isolated; never merged back without explicit promotion.
Mutable within scenario
FOUR REASONING LAYERS
Grounded
Point-in-time queries against Base + Evidence. "What does the source corpus say, character-exact?" — no interpretation, no inference.
Product surface
Used by every product to render the source-span citations a practitioner sees on hover.
Context
Typed queries against Analytical + Narrative + Commitment. "What is the structure of the case?" — actors, claims, interests, constraints, leverages, commitments, events, framings.
Product surface
PRAXIS reads CTX + EVD; CONCORDIA writes Case-scoped CTX during live deliberation.
Evidence
Provenance traversal. "For this claim, what is the chain of source spans, derivations, and revisions?" — the audit trail.
Product surface
Surfaces in inspect-edit-export-trace flows; every PRAXIS export carries the EVD trace.
Reasoning
Open-question layer. The LLM is brought back in, constrained by the typed graph as a hard scaffold rather than a soft retrieval source. Answers without traces are rejected.
Product surface
Wind Tunnel writes Scenario subgraphs; ARGUS writes Base + Evidence on ingest.
QUERY SEMANTICS
The bi-temporal model supports four canonical query forms; three are shown below. The schema is portable to any property-graph store that supports indexed temporal range queries.
As-of-system-time — what did the system know at time T?
MATCH (c:Commitment)-[r]-(o) WHERE r.t_tx_start <= $T AND (r.t_tx_end IS NULL OR r.t_tx_end > $T) RETURN c, r, o
As-of-valid-time — what held in the world at time T?
MATCH (c:Commitment)-[r]-(o) WHERE r.t_valid_start <= $T AND (r.t_valid_end IS NULL OR r.t_valid_end > $T) RETURN c, r, o
Drift query — ordered sequence of valid-time intervals
MATCH (n {id: $target_id})-[r:HELD_TRUE]->()
RETURN r.t_valid_start, r.t_valid_end, r.t_tx_start, r.t_tx_end
ORDER BY r.t_valid_start ASCENGINEERING THESIS
A naive system stores everything in one mutable graph, and pays the bill in three places: the audit trail is impossible to reconstruct because corrections destroy the prior state; cross-case extensions drift into mutually unintelligible dialects because there is no boundary at which to validate them; counterfactual analyses contaminate the production graph because there is no separate space to hold them.
Partitioning state across seven graphs with distinct mutability contracts solves all three. Base and Evidence are immutable, so the audit trail survives every revision. Analytical is append-only at the transaction-time layer, so corrections generate new edges rather than overwriting old ones. Case is forkable per dossier; Scenario is forked from Case and never merged back without explicit promotion. The boundary is architectural, which means it is enforceable.
The same partition makes the engine inheritable. A new analyst opens the Case, queries CTX for the structure, queries EVD for the chain of source spans, and inherits the reasoning rather than re-deriving it from a folder of PDFs [, ].