A bridge-inspection office shuffles report IDs into train and eval files but copies any row that mentions urgent into both so the eval looks realistic. What defect is that?
Select an answer to reveal the explanation.
Short Explanation
Copying every urgent row into both train and eval so the eval looks realistic is a data-prep leak. The same row sitting in both files contaminates the split. That is not a Domain 3 envelope peek and not a privacy-consent issue. Fix the files first.
Full Explanation
Train-set leakage into eval is a data-prep defect when the same row, or a copied urgent row, sits in both files. That is distinct from an experimenter peeking at a sealed set and from a privacy principle. Copying rows so the eval looks realistic still contaminates the split. Fix the files before anyone trains.