TACITUS

Data structures

Deterministic knowledge.
Inspectable meaning.

Define concepts, link evidence and declare the checks that matter to a policy task.

A knowledge pipeline

Prepare once.
Reuse with context.

Each stage keeps a record a reviewer can inspect. Automatic extraction and hosted agent integration are in development.

  1. Capture

    Keep the source, exact passage, date and version.

  2. Identify

    Resolve actors carefully; retain aliases and uncertainty.

  3. Qualify

    Record attribution, scope, conditions and proposed interpretations.

  4. Select

    Retrieve for the question, deduplicate records and disclose omissions.

  5. Review & reuse

    Attach a capsule, produce an artifact and check the result.

Diplomatic language / data-model demonstration

Keep the reservation
in the draft.

Words carry a position, a speaker and conditions. Model those as connected objects so a proposed summary can be checked against the original.

Original wording / fictional exercise

Opening the example…

Inspect the exact record
Read the complete fixture · JSON ↗
Authored wording and proposed annotations. This demonstrates a representation; it does not infer legal effect or make a live drafting-model call.
What is working, and what is in development?
This page loads versioned objects and shows their source wording, explicit identity, conditions and review state. The public Tacitus workbench demonstrates bounded local context reuse. Automatic extraction, expert validation and a verified bridge to hosted agent workflows remain work in progress.
Read the GitHub architecture and integration boundaries ↗
Expertise made inspectable
Capture the method,
alongside the facts.
An ontology defines the actors, statements, sources and relationships an analyst needs. A semantic layer brings the relevant parts together for one question.
Expert reasoning becomes explicit review questions, evidence requirements and counterchecks. The checklists here are Tacitus-authored frameworks for expert review. A published vocabulary or an attractive graph does not establish a case finding.
A deterministic layer
Define the concept.
Preserve the distinction.
A proposal is not a commitment. Adoption is not applicability. Give AI explicit terms, source requirements and rules it can inspect.
18 application definitions. References inform the model; meanings remain subject to context and expert review.
Opening the knowledge database…
Why this layer?
Open a layer to inspect its purpose.
Published vocabularySource materialTask & assessment
Explore nodes as a list
From a source
to a task.
An ontology defines the terms. A semantic layer connects them to a task. Inspect the graph to see what supports each connection.
1
Keep the source.
Store a dated representation, original URL, content hash and exact passage boundaries.
2
Name the meaning.
Reuse published types and relations. Keep attribution, conditions, scope and interpretation explicit.
3
Select for a task.
Retrieve relevant evidence, retain required qualifications, expose omissions and respect a budget.
4
Measure the AI input.
Verify source fidelity, then test the model’s answer, citations, total tokens and task usefulness.
Explore the database.
Search the concepts, passages and source records behind a layer. Open a result to inspect its meaning and evidence.
Use it for a task.
Open a capsule that carries the relevant sources into a specific task.
Explore capsules & artifacts
Source inspection