Usually in such competitions you need to submit predictions for a fixed number of rows.

Here, it seems like the system accepts any number of rows. E.g. the sample submission 282 rows, while there are the 3384 entries in the test file.

I would understand if the missing rows would be automatically considered incorrect but it seems to not be the case: I've tried submitting predictions for 700 test rows and truncating the same file to 290 rows and the latter had a better score, which doesn't make sense to me.

So basically how this variable # of rows in submission handled in metric calculation?