Understand
Map the question to a domain. Ask for clarification when it’s ambiguous.
A DIFFERENT FOUNDATION FOR AI
Intelligence.
With nothing
to hide.
What if you could open an answer
and see everything holding it up?
A proposal for knowledge you can trace.
Reasoning you can
inspect. Authority you can challenge.
THE CENTRAL IDEA
Every answer has roots.
You should be able to see them.
FREDERICK WOLFE · SD TECHRESEARCH PROPOSAL / AUGUST 2026
Don’t just ask
“What’s the answer?”
Ask what it
stands on.
Dominion proposes a structured knowledge system in which a response is connected to specific claims, their sources, and the principles beneath them.
You could follow those connections yourself. If the evidence changes, the knowledge can change—with a visible record of why.
This is an exploration of the whitepaper’s proposed architecture. The interactive examples are illustrations, not a live knowledge engine.
02 / THE KNOWLEDGE INSTRUMENT
An answer is the surface.
Go underneath.
Select a layer. Follow the idea
from a response to its roots.
01 / WATER DROPLETS
The answer you see.
A useful response is assembled from claims in the graph. Each factual sentence points back to the nodes that support it.
ILLUSTRATIVE ROOFING QUESTION
“Which coating is appropriate for this roof?” The response should identify the conditions and supporting evidence—not simply sound confident.
Look closer: the technical view
A Response stores sentence-to-node references, a template identifier, and the graphStateHash used at assembly. The proposed ASSEMBLED_FROM edges support recursive inspection.
SCHEMATIC 01 — A CONCEPTUAL HIERARCHY; THE UNDERLYING MODEL IS A TYPED PROPERTY GRAPH. WHITEPAPER §§2–5.
03 / INSIDE THE ENGINE
The graph owns the facts.
The language is the interface.
Trace
Find supported claims and follow their evidence paths.
Challenge
Surface conflicting positions. Keep uncertainty visible.
Assemble
Build the answer from traceable claims, with a reference for each factual sentence.
Remember
Stamp the graph state so the answer can be inspected and reproduced.
THE HONEST EMPTY STATE
“Dominion has no validated knowledge on this.”
The proposal makes room for an answer that admits its limits.
04 / IDEAS WITH THEIR BOOTS ON
Big questions.
Real-world consequences.
PROPOSED FIRST PILOT / COMMERCIAL ROOFING
The right material.
The right material.
For reasons you can inspect.
An estimator comparing roof coatings could trace a recommendation to product documentation, compatibility testing, applicable standards, and material science.
The practical gain: a new team member learns how to evaluate the decision, not just repeat it.
Illustrative future workflow. Not technical advice or a claim of current product capability.05 / AUTHORITY THAT CAN BE QUESTIONED
Expertise matters.
So does the right
to disagree.
In the proposed governance system, credibility belongs to a domain. Being an excellent physicist does not automatically make you an authority on roofing.
Challenges carry evidence. Decisions publish their reasoning. Minority opinions remain part of the record.
Who watches the experts?
The paper proposes verified identities, independent validators, time-limited panels, public scoring inputs, conflict declarations, and cross-panel appeals. These mechanisms need adversarial testing; they do not make institutional capture impossible.
-
01
Challenge a claim
Name the claim and bring evidence or a logical argument.
-
02
Open the evidence
Route the appeal, disclose conflicts, and hear both sides.
-
03
Publish the judgment
Record votes, reasons, dissent, and the resulting graph changes.
-
04
Keep the door open
Allow a further review through independent panels.
06 / UNDER THE GLASS
Enough detail
to ask harder questions.
A graph with enforceable rules
Typed nodes represent Axioms, Frameworks, Assertions, Responses, Sources, Experts, Panels, Appeals, and Versions. Support edges must be acyclic. Assertions cannot retain validated status with no citation or support.
confidence(A) = min(chainFloor, citationStrength,
validationWeight)
The weakest-link rule caps confidence. A numeric score is a proposed aggregation mechanism—not a calibrated probability unless independently validated.
Evidence that cannot quietly change
Sources are identified by a content hash. Revised bytes create a new source identity. Dependent assertions enter review. Append-only Version nodes preserve the history of changes.
A hash establishes byte identity, not the truth of a source. Sound provenance still needs sound evidence.
Reproducible answers, with conditions
The deterministic core is intended to return the same wording for the same query and graph state. Exact replay also requires pinned parser, templates, traversal rules, and software versions.
The whitepaper permits optional language smoothing. That layer must be constrained and versioned to preserve grounding and reproducibility.
The proposed reference stack
Neo4j for the knowledge graph; PostgreSQL for governance records; content-addressed object storage for source documents; a vector index for lookup; versioned APIs for queries and recursive inspection.
Identity documents remain in a separate vault. The knowledge graph holds opaque identity references.
What the pilot needs to prove
The paper scopes a commercial-roofing pilot: approximately 40 axioms, 200 frameworks, 5,000 assertions, and 1,500 sources. These are targets, not a claim of an existing corpus.
Evaluation includes sentence-level traceability, appeal completion, integrity constraints, and measured reduction in verification time.
Open questions worth working on
Who ratifies foundational premises? How should competing frameworks coexist? How is confidence calibrated? Can governance resist coordinated capture? How should private or licensed evidence be inspected?
The paper’s strong claims about eliminating hallucination and resisting censorship are architectural aspirations. Provenance and deterministic assembly alone do not establish correctness.
READING GUIDE — WHITEPAPER §§5–11 AND §15. COVER VERSION 4.0; THE SOURCE DOCUMENT RETAINS SOME EARLIER VERSION LABELS.
THE DOMINION
AI PROJECT
Hierarchical memory.
TECHNICAL WHITEPAPER
FREDERICK WOLFE / AUGUST 2026
AI PROJECT
Hierarchical memory.
Weighted credibility.
Inspectable
knowledge.
TECHNICAL WHITEPAPERFREDERICK WOLFE / AUGUST 2026
07 / GO ALL THE WAY DOWN
The whole argument.
In your hands.
Read the proposed schema, credibility engine, appeal lifecycle, inference pipeline, and pilot plan in the original Dominion AI whitepaper.
Download access is included with a successful subscription to the Dominion app.
Explore Dominion & subscribe ↗The subscription supports the existing Dominion app. The research architecture described here is a separate proposal.
08 / A GOOD IDEA WELCOMES A GOOD QUESTION
Bring your curiosity.
Or your best objection.
Domain expert, engineer, researcher, potential pilot partner—or simply someone who sees a better way. Let’s hear it.
Your message is delivered privately to the project’s inbox. Contact details are never displayed on this site. Please don’t include confidential or sensitive information.