A courthouse FAQ set was split by filename. Many eval questions are the train questions with a comma moved, and scores look perfect. What is the issue?
Select an answer to reveal the explanation.
Short Explanation
Eval questions that are the train questions with a comma moved are near-duplicate overlap. Perfect scores are a dataset defect, not a brilliant model. Deduplicate the files. That is not a Triton green light, an envelope-peek rule, or a comma-moving kernel.
Full Explanation
Overlap and near-duplicates across train and eval files are a dataset defect. A comma-moved FAQ that sits in both splits inflates scores without proving skill. That is not a serving green light and not the same as peeking at a sealed envelope. Deduplicate the files before anyone celebrates the score.