Model-vs-data metrics for every pair in a study

RMSE, MAE and max residual per pair, each computed over every sample.

The report's table shows these for all pairs at once, which is why they are served apart from the channels: downloading both full datasets per pair just to fill three numbers in a row would be wasteful.

A pair whose simulation has no results is omitted rather than failing the batch — its row renders as pending, which is what it is.