We made a fine-grained human annotation dataset for simplification by LLMs, one of the largest scales to date.