In short: If a problem has 4 sample test cases, and code fails at test 4 then wrong submission is counted, but if it fails at test 1, this code is not counted as wrong submission.
We often over-read the sample output and submit the code, which costs us points. I suggest, if any of the sample test cases fail, wrong submission shouldn't be counted.
MikeMirzayanov, Please look into it.