To the experienced data scientists / statisticians here: does this sort of regression analysis between the actual and predicted age seem legit to you? It seems

To the experienced data scientists / statisticians here: does this sort of regression analysis between the actual and predicted age seem legit to you? It seems very weird to me. From what I understand, the p-value tests the hypothesis that the predicted and actual data are correlated, which seems like a very weak way of assessing the reliability of the prediction. An estimator of the test error seems more relevant, and that's kind of what the R^2 does, but why do an affine regression between predicted and actual value instead of say an RMS error? Isn't this using the test data for parameter-fitting, i.e training?