> ## Documentation index
> Fetch the complete documentation index at: https://docs.prescientai.com/llms.txt
> Use this file to discover all available pages before exploring further. Add `.md` to any page URL to get that page as Markdown.

# Configs and the Preference Score

> What a model config is, what Default means, and how to read the Preference Score and the criteria that Model Comparison shows for two configs.

A model can have more than one config. A config is one configuration of a model, and each config has its own results. Model Comparison puts two configs side by side and gives each one a **Preference Score**.

In a label on this page, a word in braces is a value from your account. The efficiency is the efficiency metric of the target, for example ROAS. The prefix is a letter that comes from the target.

## What is the Default config?

Each model card on the overview shows one config. When the model has a config with the name **Default**, the card shows that config. The **Avg Accuracy** tile also uses the **Default** config of each model.

A card shows **Config** when the model has one config, and **Configs** with a count when the model has more. On a detail page, the config selector lists the configs with **Default** first.

> **You cannot add or change a config in Model Center:** The names of the other configs come from your account. Model Center has no control to add a config or to make a different config the default.

## What is the Preference Score?

The **Preference Score** is a number from 0 to 100 that Model Comparison gives to each of the two configs. The product says that the score measures model quality across different criteria.

The config that the comparison prefers has the **Recommended** badge. The result under the two panels reads **Model A Preferred**, **Model B Preferred**, or **Equally Preferred**.

The score is a summary. The rows under **Additional Metrics** show the criteria behind it.

## What do the criteria mean?

Each panel lists four criteria. The names of two criteria start with a letter and **-MMM**. The letter comes from the target, for example R for a revenue target.

| Criterion            | Plain meaning                                                                                                    | Better value                     |
| -------------------- | ---------------------------------------------------------------------------------------------------------------- | -------------------------------- |
| Fit                  | How well the model fits the reported total.                                                                      | Nearer to 100                    |
| Expectedness         | How typical the efficiency values of the model are, compared with an industry average.                           | Nearer to 1                      |
| Outliers             | Whether the model has an efficiency value far outside the normal range for a channel.                            | **No outliers**                  |
| Paid % within bounds | The paid percentage of the model. The product calls a value roughly above 30% and not more than 100% reasonable. | Above 30% and not more than 100% |

Expectedness is the least familiar of the four. A value near 1 means that the results of the model are standard and expected. A value near 0 means that the results are unusual.

A criterion can show **NA** in place of a number.

For the tooltip that the product gives for the score and for each criterion, see [Tooltips of the criteria](https://docs.prescientai.com/reference/model-center/model-comparison#tooltips-of-the-criteria).

## What do the colors of the criteria mean?

Each value has a color that shows where the value falls. For example, the fit value is green at 85 or more. For each color and its limits, see [Colors of the values](https://docs.prescientai.com/reference/model-center/model-comparison#colors-of-the-values).

## Which run does a comparison describe?

A comparison uses one run of each config. Each panel shows the date of that run as **Run compared**.

If a config has a later run, the panel also shows **Latest run** with its date. The figures of the comparison still describe the earlier run. To read the latest run of a config, open the config in Model Center.

## Next steps

- [Compare two configs](https://docs.prescientai.com/model-center/compare-two-configs): open Model Comparison and read the result.
- [Accuracy and fit](https://docs.prescientai.com/model-center/accuracy-and-fit): the two scores of a single model.