Decided to consider defensive criteria to avoid collecting malicious data during the data‑collection process
Since AI‑generated code or documents may contain ASCII‑code transformations, Unicode‑confusing characters, invisible control characters, obfuscated strings, etc., we need to think about how to detect and normalize them at the collection stage
Concluded that establishing criteria for filtering malicious/contaminated/obfuscated data is more important than simply gathering a large amount of GitHub data
Work may be delayed for a while due to a condition issue