Before you begin
This full local form uses 50 Reading items, 47 Listening transcript items, 12 Writing tasks, and 11 Speaking drafts, matching the current public task-count blueprint while remaining fixed rather than adaptive. Audio is represented by original transcripts. Reading and Listening remain raw accuracy measures; each of the two open Writing tasks receives one local-AI holistic 0–5 task score using the public rubric, while the current 1–6 section conversion is not inferred from this fixed form. Speaking drafts remain local and unscored.