# Knowledge Box

```{figure} ../../_static/images/tetrad-interface/box-by-box/knowledge-box.png
:name: tetrad-knowledge-box-screenshot
:alt: Knowledge Box in the Tetrad interface.

Knowledge Box in the Tetrad interface sidebar and main panel.
```

## Purpose

The **Knowledge** box is where you specify **background knowledge** about the causal structure that
should constrain search algorithms and other procedures in Tetrad. This knowledge can include:

- **Forbidden edges** – edges that must **not** appear (e.g., `Y → X` is disallowed).
- **Required edges** – edges that **must** appear if compatible with the graph type (e.g., `X → Y`).
- **Tiers or temporal ordering** – variables that are earlier or later in time, so edges cannot go “backwards”.
- Other structural constraints, depending on the algorithms and version.

You use this box to:

- Encode strong domain knowledge (e.g., time ordering, interventions, biological directions).
- Rule out implausible models.
- Improve both correctness and efficiency of search.

## Typical workflow

1. **Identify variables and constraints**
   - After loading data in the *Data* box (and possibly defining graphs), decide:
     - Which variables clearly precede others (e.g., baseline covariates before outcomes).
     - Which directions of influence are impossible or mandatory.
   - Open the Knowledge box to create or edit a knowledge object.

2. **Define tiers (optional but common)**
   - In the tier editor:
     - Place variables into ordered blocks (tiers) such that edges are only allowed from
       earlier tiers to later tiers.
   - This is useful for:
     - Longitudinal data (time 1 → time 2 → time 3).
     - Clearly staged processes (e.g., genotype → phenotype → outcome).

3. **Specify forbidden and required edges**
   - Use the edge constraints editor to:
     - Mark specific directions as **forbidden** (e.g., `Y → X` cannot occur).
     - Mark specific directions as **required** (e.g., `X → Y` must occur, if the graph
       type allows it).
   - You can often select variables from drop-downs and click buttons to add the constraint.

4. **Use knowledge in search**
   - In the *Search* box, select the knowledge object you defined.
   - When you run a search algorithm that supports background knowledge:
     - Forbidden edges will never be suggested.
     - Required edges will be enforced where compatible.
     - Tiers will restrict possible edge directions.

5. **Refine and reuse**
   - After seeing search results:
     - You may update knowledge (e.g., adding new constraints suggested by domain experts).
   - Save the project to keep the knowledge object for future runs or related datasets.

## Key controls

- **Toolbar**
  - **New** – create a new knowledge object.
  - **Duplicate / Rename / Delete** – manage existing knowledge objects.
  - **Export / Import** – save or load knowledge from files (when supported).

- **Knowledge list**
  - Shows all knowledge objects defined in the project.
  - Selecting one loads its tiers and edge constraints into the main panel.

- **Main panel**
  - Typically divided into:
    - **Tiers editor** – assign variables to ordered tiers.
    - **Forbidden / Required edges editor** – specify edge-level constraints.
  - May include additional options for:
    - Allowing or disallowing certain graph types or patterns.
    - Interpreting knowledge in different ways for different algorithms (depending on version).

## Common patterns & tips

- Start with **coarse-grained knowledge** (e.g., tiering) before adding many detailed edge constraints.
  - Tiering is easy to specify and often very powerful.
- Only add **strong, defensible constraints**:
  - Overly aggressive constraints can force algorithms to miss valid models.
  - Under-specified knowledge is safer but may give less benefit.
- Keep separate knowledge objects for:
  - Different experimental conditions or subpopulations.
  - Different modeling assumptions (e.g., conservative vs. aggressive prior constraints).

## Related pages

- `Tetrad Interface → Overview` – high-level tour of the GUI.
- Other boxes that commonly interact with **Knowledge**:
  - *Search* (uses knowledge to constrain causal discovery).
  - *Graph* (resulting graphs respect the knowledge, where supported).
  - *Simulation* (you may design simulations that respect a particular knowledge structure).