Which metrics to use for model evaluation ?

[George3d6]

Hello, thank you for maintaining this library, I find it takes away one of the most daunting tasks in regards to tertiary structure predictions for those that aren't that familiar with the field.

However, there's one thing that I can't figure out and reading through CASP related material (e.g. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4394854/) doesn't really help, how do you actually validate the models ?

As in, once you have a model that seems to predicted scaled values for the 3x3 matrices well under a given loss function that's practical for optimizing the model, how do you determine if the predicted structures would be correct for domain specific purposes ?

Assuming said evaluation is extremely hard is there some simpler metric that gets you 9x% of the way there so that you can at least tell if your model is heading in the right direction ?