The mathematics of why.
Correlation-based AI, including every large language model, learns the patterns in historical data. Decimitry extracts and proves the mechanisms that produced it. That distinction sounds philosophical. In practice, it is the difference between a system that describes your business and one that can reason about changing it.
Causal mechanism discovery
Every enterprise dataset is the shadow of a causal system: prices, operations, demand, and risk pushing on each other through real mechanisms. Standard ML fits the shadow. Statistical causal inference goes a step further and estimates which variables drive which. Decimitry does not stop there: we extract the underlying mechanisms and prove them mathematically, so each one holds as an engineered component, not a statistical guess.
Because the model is built from proven mechanisms, it answers questions pattern-matching cannot: What is actually driving this cost line? Which lever moves this outcome, and which merely co-moves with it? What happens under conditions we have never observed?
Causal inference is the ceiling for statistical approaches. For Decimitry, it is the floor.
Continuous-time modelling
Real systems evolve continuously; data arrives in fragments. Most AI is built for the fragments: discrete snapshots, fixed intervals, big batches. Decimitry embeds your data into calculus models that are continuous in time: differential equations whose structure carries information between the data points.
The practical consequence is data efficiency. Where deep learning needs millions of examples, a mechanism written as mathematics needs only enough data to calibrate it. Our models operate in low- and minimal-data environments: new markets, rare events, systems that have never been instrumented.
Counterfactual simulation
A model built from proven, continuous-time mechanisms can be run forward under conditions that never happened. Raise the price. Reroute the supply line. Change the policy. The model propagates the intervention through every causal pathway and returns the trajectory, alongside the counterfactual trajectory of doing nothing.
Decisions stop being bets on a forecast and become comparisons between simulated futures, each with quantified consequences and explicit, inspectable assumptions.
Every mechanism we prove is added to the Causal Genome, Decimitry's growing library of cause-and-effect machinery, and made callable through deterministic Decision APIs.
Pattern AI vs. proven mechanisms
| Correlation-based AI | Decimitry | |
|---|---|---|
| Question answered | What comes next? | Why, and what happens if we act? |
| Treatment of time | Discrete snapshots | Continuous trajectories |
| Data required | Massive historical corpora | Works in low / minimal-data settings |
| Under regime change | Degrades silently | Mechanisms persist; model adapts |
| Explainability | Post-hoc approximations | Proven: the mechanism is the explanation |
| Interventions | Out of scope | The core operation |
Oxford applied mathematics, applied.
Decimitry's methods descend from the Oxford school of applied and industrial mathematics. Our scientific advisors, Prof. John Ockendon FRS, IMA Gold Medallist and a founder of the Oxford Centre for Industrial and Applied Mathematics, and Dr Hilary Ockendon, have spent five decades turning differential equations into industrial results.