Use the built-in review tools rather than editing exports: approvals to catch bad responses before they count, the deduplicator to merge sites recorded twice, and validation in the form to stop the error recurring. The corrections stay in the system, so the data everyone reports from is the corrected data.
Video: Cleaning Your Data in mWater on YouTube
Cleaning in a downloaded spreadsheet is the most common and most damaging shortcut in monitoring work. It creates a private, better copy of the data while the shared system keeps serving the bad version, and it fixes symptoms without touching the form that produced them. Cleaning in place is slower to start and much cheaper over a programme's life, because each fix is visible and each cause can be closed off.