We publish our standard. Every claim is traceable to the theory that generated it. Every output has a cause.
Every model can be challenged - and will survive the challenge.
Not rules we follow. Commitments we made when we decided to build the right way.
01
Data tells you what happened. Theory tells you why.
Most systems find patterns in data and call it intelligence. We build formal models of human behaviour - grounded in decades of social science, cognitive science, and complexity theory - and use data to validate them.
The difference is not technical. It is epistemological. A pattern without a cause is not an explanation. We do not sell unexplained patterns.
02
We do not produce probabilities. We produce traceable conclusions.
Every output from our platforms can be followed back to a specific theoretical mechanism. You can ask why - and get an answer that is not a confidence score, but a cause.
This is what we mean by glass boxes, not black boxes. If we cannot explain it, we do not ship it.
03
A model that has never been tested against reality is not a model. It is a hypothesis.
We validate at five levels - from statistical fidelity to reconstructive validation against known real-world outcomes.
Most systems stop at level one. We do not consider that sufficient.
04
We hold ourselves to a standard most do not publish. Not because it is required. Because it is the only way to build something you can defend - in front of a client, a regulator, or a scientist.
Every limitation is documented. Every assumption is named. Every validation gap is disclosed. The platform pages carry the current validation status for each service.
Epistemic honesty is not a weakness. It is the moat.
Statistical fidelity - distributions that match - is necessary. We do not consider it sufficient.
A model that reproduces marginals but not mechanisms is a deepfake of social science. We call it statistical mimicry. We do not build it. We do not sell it.
Reconstructive Validation
The model reproduces known, documented real-world outcomes. This is the gold standard. This is where we build.
Mechanistic Grounding
Every output traces to a specific causal mechanism derived from theory. Not a probability. A cause.
Behavioral Plausibility
Domain experts - sociologists, cognitive scientists, complexity researchers - recognize the outputs as realistic.
Structural Validity
Relational patterns and network topology are preserved across the model.
Statistical Fidelity
Distributions match known empirical data. Necessary. Never sufficient alone.
We build to L5. That is the difference between correlation and cause - between data that looks right and data you can defend.
"Rigour is not the enemy of profit. Rigour is the profit." - Prof. Finch, Quantumeon Chief Scientific Advisor
If you cannot trace a prediction to a cause, you cannot defend it. We can always trace. That is not a feature - that is the requirement we set for ourselves before we wrote a line of code.
Outputs a probability. No mechanism provided.
Cannot explain why the model says what it says.
Distributions may match. Causes are absent.
Indefensible in front of a regulator or scientist.
Statistical mimicry dressed as intelligence.
Every output traces to a specific theoretical mechanism.
You can ask why - and receive a cause, not a score.
Validated at five levels. Validation level disclosed on every report.
Defensible in front of a client, a regulator, and a scientist.
Mechanistic transparency as a product requirement, not a bonus.
Every product, every feature, every output must pass three independent gates before it reaches a client. Not one. Not two. All three.
"Is the methodology defensible in a peer-reviewed journal?"
Fails if: No theoretical grounding. Unfalsifiable claims. Black-box generation.
"Would we be comfortable if the population being modeled saw how we modeled them?"
Fails if: Embedded stereotypes. Bias amplification. Dehumanizing reductions. Privacy violations.
"Does this create defensible, recurring value that clients will pay a premium for?"
Fails if: Commodity feature. No moat. Race-to-bottom pricing.
Scientific rigour, ethical integrity, and commercial defensibility are not in tension.
They are the same thing. They are the moat.
The same four principles and five validation levels apply across every platform - adapted to the science each one compiles.
How habitus, capital, and field - Bourdieu's core concepts - get formalized into agent specifications. The translation from sociology to computation, step by step.
L1 through L5 explained. Why statistical fidelity is the floor, not the ceiling - and what mechanistic grounding actually means in practice.
This is our standard. Three platforms, thirty services, five levels of validation - all governed by these principles. Now hold us to it.
