A question about fine tuning statistical data

Really just a low priority question, if someone can share info on this.
When the sample size in the training file is large, the statistical data provided by the training is for every 10th sample. Can someone confirm that it is just a sampling of every 10th item (which is overall still a pretty good indication of the training results) or it is an average of the 10 items (1 to 10, 11 to 20, 21 to 30) which is a more accurate statistical data than just single samples.
thank you