Build a Rejection Ladder

Create four disposable inputs failing at different boundaries: size, encoding, CSV structure and minutes range. Implement parsing and predict where each fails. Keep one valid empty file and one repeated-row file to distinguish rejection from valid edge cases. Evidence: input bytes or fixture names, expected stage and actual exception/result. Pass when invalid records cannot produce a successful partial report.

Keep predictions before observations. Use the guided lab for commands and contracts; use calibration after your attempt.