Documentation

How to get from a dataset to a causal graph you can act on.

Start here

Three steps to your first causal graph

01

Create an account

02

Upload a dataset

Bring a CSV or JSON file (save Excel sheets as CSV first). Columns become candidate causes and effects — no labelling needed.

03

Run discovery

Pick an algorithm and threshold, then read the graph: every edge carries a signed weight, and a confidence score when the algorithm measures one.

Run discovery from your own code

Post a table of observations and get back the edges the engine could defend, each with a signed weight, and a confidence score when the algorithm measures one.

curl -X POST "$CESTRUM_API_URL/api/causal/run" \
  -H "X-Api-Key: $CESTRUM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "datasetName": "pricing",
    "modelType": "PC",
    "variableNames": ["price", "demand", "inventory"],
    "data": [
      [12.5, 840, 1200],
      [11.0, 910, 1040]
    ]
  }'

Guides

Getting Started

Create an account, upload your first dataset and run a causal discovery. The workflow goes from data upload to an interactive causal graph you can explore, test and export.

Causal Discovery Guide

How UCMT discovers causal structure from observational data: how to read an edge’s signed weight and, where the algorithm measured one, its confidence, and how to check results against what you already know about the domain.

Inference & Interventions

How to test an intervention on a discovered graph: choose the variable to change, set its value, and read the estimated effect on the variables downstream of it.

Dataset Best Practices

Guidelines for preparing data for causal discovery: handling missing values, encoding categorical variables, and making sure you have enough rows for the relationships you want to identify.

API Integration Guide

Calling the platform programmatically: authenticate, upload a dataset, start a discovery run and fetch the resulting graph. The API reference lists the endpoints.

Reference

User guideHow each screen of the application works, with screenshots API ReferenceThe REST endpoints behind the application

Something not covered here?

Send us the dataset shape and what you expected — we answer technical questions directly.