# Detail: Generalized Instantiated Model

This page describes **Generalized instantiated models** in the
**Instantiated Model** box. These are **custom models with user-specified
functional forms and error distributions that have been fitted (or at least
evaluated) on data**, starting from a Generalized parametric model.

```{figure} ../../_static/images/tetrad-interface/box-by-box/generalized-im.png
:name: tetrad-beneralized-im-screenshot
:alt: Generalized Estimator

Generalized Estimator
```

A Generalized instantiated model contains:

- The underlying **graph structure**.
- Concrete parameter values for the **specified functions** at each node.
- Any **error-distribution parameters** used in the model.
- Optional fit summaries or scores, depending on the estimator.

## How Generalized instantiated models are created

1. In the **Parametric Model** box, build a **Generalized** model:
   - Specify the functional form for each variable given its parents.
   - Specify the error distribution for each variable.
2. In the **Estimator** box (if supported for your Generalized setup), select:
   - The Generalized parametric model, and
   - A dataset.
3. Run an estimator or evaluation routine to obtain parameter values and
   fit summaries.
4. Save or send the result to the **Instantiated Model** box.

In some workflows, Generalized models are used mainly for simulation and are
“instantiated” by construction rather than by fitting.

## Instantiated Model box layout (Generalized)

When you select a Generalized instantiated model, the main panel typically
includes:

- A summary of the **functional form and parameters** for each variable.
- Any **estimated error-distribution parameters**.
- Fit or evaluation metrics, if the estimator computes them.

Because Generalized models are highly customizable, the exact layout may vary
more than for the other model families.

## Typical uses

Generalized instantiated models are useful when you want to:

- **Simulate complex data** (nonlinear, non-Gaussian) that reflect a particular
  causal story, then test search algorithms against it.
- Evaluate whether a proposed nonstandard model provides a better fit than
  simpler Bayes/SEM/Hybrid models.
- Store and document the exact parameterization used in a simulation study.

## Tips

- Document your choices of functional forms and error distributions carefully,
  for example using a *Note* box and meaningful model names.
- Start with a simpler special case (e.g., linear with non-Gaussian errors)
  and then add complexity, so that you can debug estimation and simulation in
  stages.