I was working during the last year on developing an approach that determines human age class (young or old), but after I got exciting results I've discovered that I made a mistake when implementing some lines of code. I think that the mistake is a bit critical because I was comparing my approach and literature's ones. The mistake I made was during the classification task, where the data was entered in the right way but the experience was badly managed.
What should I do? Do I have to stay quiet or do I have to fix it? Note that the paper is not yet published nor peer reviewed.