> For the complete documentation index, see [llms.txt](https://docs.rootcause.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.rootcause.ai/more-details/digital-twin/model-comparison.md).

# Model Comparison

Causal models evolve. You might train a Digital Twin on January data, then retrain on February data. Or experiment with different configurations — omitting certain variables, adding domain knowledge, changing temporal settings.

Comparing versions is how you tell what a retrain actually did: which relationships appeared, disappeared, or reversed direction, and whether predictions got better or worse.

This is essential for model governance. Before promoting a new model version, you need evidence of how it differs from the current one.

There is no separate Compare page. Comparison happens in three places on the twin itself, and each answers a different question.

***

### Comparing metrics across versions

The twin's **Evaluation** tab compares versions of *that twin*. Its right-hand **Configuration** panel carries a **Version comparison** multiselect: tick one or more earlier versions and every chart and per-variable row on the left gains a series for each, so you can see whether a change made predictions better or worse.

Each selected version appears as its own bar colour and its own indented table row. This is the fastest way to answer "did the retrain help?", variable by variable rather than as one averaged number.

***

### Seeing what a version changed

[Version History](/more-details/digital-twin/version-history.md) lists every version of the twin with its status, creation time, type and model fit. Select a version and a panel opens showing what changed relative to the version it was derived from: the temporal dependencies and the known and blocked constraints that were edited.

Use this to see which configuration edits produced a version, and to switch the active version or roll back. For the structural before and after — relationships added, removed or reversed — compare the graphs themselves.

***

### Comparing environment groups

On a multi-environment (panel) twin, the scope selector's **Compare** toggle marks environment groups of the same twin for comparison — up to four — and merges the graphs of the two drawn on the canvas into one diff. The legend colour-codes edges present in only one group, present in both, or pointing the opposite way between them.

This compares groups within one twin, not two twins or two versions.

***

### Interpreting differences

**Structural changes**

* **New edges** — the model found a relationship not present before; may reflect new patterns in updated data
* **Removed edges** — a previously discovered relationship is no longer supported; may indicate data drift
* **Direction changes** — causality now flows the opposite way; investigate why

**Questions to ask**

1. Are structural changes expected given the data differences?
2. Do direction changes make domain sense?
3. Would these changes affect downstream decisions?

***

### Comparing two configurations

To try a different configuration, open **Modify model** on the twin, change what you need and run. That produces a new version of the same twin, which the Evaluation tab and Version History will then compare against the ones before it.

Building a second, separate twin does not give you a cross-twin comparison: nothing in the product compares two different twins.

For Digital Twin creation steps, see [Build Digital Twin](/user-guide/creating-digital-twin.md).