Anonymous Dataset for Semi-Structured Extraction under OCR and Annotation Noise

This dataset is released for anonymous peer review. It contains OCR-derived clinical report texts with synthetic noise injection and corresponding key-value annotations. All identifying information has been removed.