Cutting-edge research, made accessible.
Relevance-Based Prediction (RBP) is novel mathematical research developed by the founders of Cambridge Sports Analytics. RBP treats every prediction as an information problem — weighting historical observations by how much they actually inform the present.
Most prediction systems hand you a number. Our method tells the story behind every prediction.
By grounding forecasts in information theory, we surface which observations drove the result, how much each one mattered, and where the model is least confident.
No black boxes. Just transparent math, served at production speed.
Forecasts derived from relevance weighting grounded in formal information theory.
Every prediction comes with evidence behind it. Understand the story behind every prediction.
Designed for production from day one — three deployment shapes that match how real teams work.
A straightforward REST API with a first-class Python client. Submit predictions, pull results and analyze your results all within your existing stack.
A shippable Docker image of the full Prediction Engine. Deploy inside your VPC or on bare metal. Built for teams who need privacy, low latency, or raw throughput at scale.
An MCP server that turns the Prediction Engine into a conversation. Non-technical users can design experiments, submit predictions, and analyze results — all in natural language.
The Python SDK is designed to allow users to focus on experiment design and analysis. Pass features, get back predictions with full attribution — which historical observations mattered, and by how much.
View full documentationfrom csa_prediction_engine import predict_grid, GridOptions, PredictionResults # Submit y, X, theta — predictions come back with full attribution yhat, output_details = predict_grid(y=y, X=X, theta=theta, options=GridOptions()) # Every insight packaged in one results object Results = PredictionResults(output_details)