The future of a system depends on the path it took.
We build working models of real systems — markets, consumer demand, industrial operations — and simulate where they go next. Each forecast carries its own uncertainty and is checked against held-out reality. The result runs in production, and we keep it there.
Start a conversation →Why we exist
Language models don't do math.
The current wave of AI is superb at narration and weak at numbers. When the question is quantitative — what will demand do next quarter, how does a shock propagate through an operation, which scenario deserves capital — the answer requires real machinery: feature engineering, system identification, neural forecasting, stochastic simulation.
That machinery is what we build.
What we build
Three kinds of model.
Forecasting engines
Neural time-series forecasting at production scale, with honest baselines, walk-forward validation, and calibrated uncertainty.
Simulation & digital twins
System-dynamics and agent-based models of operations, markets, and consumers. Pose a what-if and see the range of outcomes it produces.
Decision interfaces
Scientific visualization that makes a model legible to the people who act on it — a dashboard you can read at a glance.
Method
Built one-of-one.
Each model is designed from scratch for one client and one system, then handed over as a system you own and we maintain as the data shifts. It ships only after it beats honest baselines on data it has never seen. The person who scopes the model is the person who builds it.
Work
A production track record.
The practice behind the firm has shipped forecasting and simulation that production operations depended on. Demonstrations run only on data you can audit — open sources or fully disclosed synthetic systems — with pipelines, baselines, and metrics in the open.
Contact
Start a conversation.
Write to contact@pathdependenceresearch.ai, or see what a useful first note includes.