idea_world_labDEV JOURNAL
Saturday, June 20, 2026

June 20, 2026

  • 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