Whether the validated model and the complete model are cross-tested

■ Overview

I would like to use the "Export Model" function of the Classification Learner app to predict new data.

When I read the official MATLAB documentation, it says, "Training a model in a classification learner consists of two parts."

[Accuracy (Verification)] displayed on the classification learner app is trained in the model (here, the ensemble bagging tree is used as an example) →Verification (here, cross-validation)-> Results after verification (displayed on the classification learner app), while the model exported to the workspace with a complete model (export model) is "verified". It states that you can export "models that did not".

・ Learning the classification model in the classification learner app

https://jp.mathworks.com/help/stats/train-classification-models-in-classification-learner-app.html

The main question here is

(Question 1)

Why is the exported model performing cross-validation in the learning displayed on the classification learner app?

Is there a difference that the unverified model is exported?

(Question 2)

Does it mean that if you use the exported model as it is and predict new data, the output result will be overfitted?

If the above guess is correct, is it better to perform cross-validation on the export model as well to predict new data?