Configuration
The Configuration panel defines what goes into a Digital Twin: the data it runs on, the variables included in causal discovery, the algorithms used, and any prior knowledge about which relationships are known or impossible. Get this right and discovery has a fair chance. Get it wrong and you'll miss real effects or fabricate spurious ones.
The panel is read-only by default. Click Edit & Create New Version at the top right to make changes; every save creates a new version of the twin and retrains.
For context, see Step 5: Build Digital Twin and Exploring the Causal Model.
Opening Configuration
In the Overview panel's Config section, click the View Config link. The graph stays visible on the left; the right panel switches to the Configuration view.
Data View
The dataset the twin is built from. Changing it triggers a new major version, because the twin is now based on different data.
Fields
Every column from the Data View, with its data type (Boolean, Category, Number, Date/Time, Array/Set). The header shows the include count ("20 included, 1 excluded" in the example). Use the filter at the top to find a field by name; toggle each field to include or exclude.
Exclude:
Unique identifiers (customer_id, order_id) — no causal meaning.
Metadata (created_at, updated_by) — usually irrelevant.
Redundant fields that duplicate another column.
Include anything you might want to predict, simulate, or trace a causal path through.
Algorithm Versions
Two algorithm choices, one per stage:
Causal Discovery — builds the graph. Default: SPARC (adaptive multi-stage).
Digital Twin — fits the probabilistic model on top. Default: CausalBART (Bayesian network simulation).
Variable Roles
Variable Roles fence off the ends of the causal chain:
Sources (exogenous) — pure drivers. Only other sources are allowed to influence them.
Targets (outcomes) — pure outcomes. They can only send edges to other targets.
Variables not marked are free to be cause or effect, in either direction. In the example, customerID, InternetService, MultipleLines, and PhoneService are sources; Churn is the target.
Prior Knowledge
Encode certainty about specific edges before discovery runs:
Known relationships — edges that must exist. Use when you have experimental evidence or domain certainty.
Blocked relationships — edges that cannot exist. Use when a connection is logically impossible.
Both lists are optional; both improve the resulting graph when you have real prior knowledge to add.
Variable Ordering
Constrains causal precedence at the variable level. "TechSupport must come after InternetService" tells discovery that TechSupport cannot be a cause of InternetService — only an effect, direct or indirect. Useful when temporal order is known or domain logic is clear.
What happens when you save
Click Edit & Create New Version, change what you need, save. A new version of the twin is created and causal discovery reruns against the updated configuration. The form warns whether your edit will produce a new major version (Data View change) or a minor version (everything else).
Other Working with a Digital Twin pages
Exploring the Causal Model — graph layouts and variable details.
Reviewing Model Quality — predictive accuracy and per-variable metrics.
Inspecting Causal Relationships — individual edges and their statistics.
Version History — multiple versions of the same twin.
See Digital Twin overview — general overview.
Last updated