# Relationships Tab

The Relationships tab is where you see—and refine—the causal structure your Digital Twin discovered. It's a comprehensive view of every cause-and-effect relationship in your model, with tools to edit, resolve ambiguities, and ensure the graph accurately reflects reality.

Causal discovery algorithms do heavy lifting, but they're not omniscient. Sometimes they can't determine which direction a relationship goes. Sometimes they find spurious connections. The Relationships tab is where human judgment meets algorithmic output.

(SCREENSHOT: Relationships tab showing the full list of causal relationships with editing controls)

***

### The Relationship List

The main view shows all discovered causal relationships:

**Columns:**

| Column    | Description                            |
| --------- | -------------------------------------- |
| Source    | The causing variable                   |
| Target    | The affected variable                  |
| Strength  | How strong the relationship is (0-1)   |
| Direction | Arrow showing causal direction         |
| Actions   | Edit, flip, or remove the relationship |

**Searching**

Use the search box to filter relationships by variable name. Useful when you have many variables and want to focus on specific ones.

(SCREENSHOT: Relationship list with search filtering applied)

***

### Directional vs. Bidirectional Edges

Causal discovery produces two types of edges:

**Directional Edges (A → B)**

The algorithm is confident that A causes B. These edges are ready to use.

**Bidirectional Edges (A ↔ B)**

The algorithm knows A and B are related but couldn't determine which causes which. This happens when:

* The data doesn't provide enough information to distinguish direction
* Both directions are statistically plausible
* The variables are symmetric in their relationship

Bidirectional edges must be resolved before the model can be fully trained.

(SCREENSHOT: Mixed list showing directional and bidirectional edges with different indicators)

***

### CPDAG Resolution

When your graph contains bidirectional edges, the model is in a "CPDAG" state (Completed Partially Directed Acyclic Graph). The Relationships tab helps you resolve these.

**Resolution Panel**

Unresolved edges appear in a dedicated section:

1. For each pair, decide which direction makes sense
2. Click the arrow button to set the direction
3. Repeat for all unresolved edges

**Making Direction Decisions:**

Consider:

* **Temporal order** – Does one typically happen before the other?
* **Domain knowledge** – Which direction makes business/physical sense?
* **Mechanism** – Can you explain how A would cause B, or B would cause A?

**Example:**

If you see `marketing_spend ↔ brand_awareness`, you'd likely resolve it as `marketing_spend → brand_awareness` because spending on marketing leads to awareness, not vice versa.

(SCREENSHOT: CPDAG resolver showing unresolved edges with direction buttons)

***

### Editing Relationships

Beyond resolution, you can modify the graph directly:

**Adding a Relationship**

1. Click **Add Relationship**
2. Select source variable
3. Select target variable
4. The edge is added to the graph

Use this when you know a relationship exists that the algorithm missed.

**Removing a Relationship**

1. Find the relationship in the list
2. Click the delete icon
3. Confirm removal

Use this when a discovered relationship is spurious or doesn't make domain sense.

**Flipping Direction**

1. Find the relationship
2. Click the flip icon
3. The direction reverses (A → B becomes B → A)

Use this when the algorithm got the direction wrong.

(SCREENSHOT: Relationship editing controls with add, delete, and flip buttons)

***

### Cycle Detection

Causal graphs must be acyclic—you can't have A → B → C → A. The Relationships tab detects and highlights cycles:

**Cycle Warning**

If your edits create a cycle, a warning appears showing:

* Which relationships form the cycle
* Which edges need to change to break it

**Resolving Cycles**

1. Identify which relationship is incorrect
2. Remove or flip it to break the cycle
3. The warning disappears when the graph is acyclic

(SCREENSHOT: Cycle detection warning showing problematic edges)

***

### Saving Changes

When you modify relationships:

**Creating a New Version**

Major edits create a new version:

1. Make your changes
2. Click **Save Changes**
3. A new version is created with your modifications
4. The model retrains on the new graph

**What Gets Versioned:**

* Added relationships
* Removed relationships
* Direction changes
* CPDAG resolutions

(SCREENSHOT: Save changes button with version creation confirmation)

***

### Graph Visualization

While the list view is primary, you can also visualize the graph:

**Graph View Toggle**

Switch between list and graph views to see relationships spatially. The graph view shows:

* Nodes as variables
* Edges as causal connections
* Directions as arrows

**Interacting with the Graph**

* Click nodes to see their connections
* Hover edges to see strength
* Drag to rearrange layout

(SCREENSHOT: Graph visualization showing nodes and directed edges)

***

### Best Practices

**Resolve Systematically**

Work through unresolved edges methodically. Don't skip around—you might create cycles.

**Involve Domain Experts**

Share the relationship list with people who understand the business domain. They can often quickly identify wrong directions.

**Document Decisions**

When you flip or add relationships based on domain knowledge, note why. This helps future users understand the model.

**Verify After Changes**

After making edits, check the [Evaluation Tab](https://docs.rootcause.ai/user-guide/digital-twin/tabs/evaluation-tab) to see if model quality improved or degraded.

***

### Next Steps

Once relationships are resolved and verified:

* Check the [Path Analysis Tab](https://docs.rootcause.ai/user-guide/digital-twin/tabs/path-analysis-tab) to see causal flows
* Review the [Evaluation Tab](https://docs.rootcause.ai/user-guide/digital-twin/tabs/evaluation-tab) for model performance
* Run [Simulations](https://docs.rootcause.ai/user-guide/digital-twin/tabs/simulation-tab) to test scenarios