# Four-record technical report

The study references the MIT-BIH Arrhythmia Database, PhysioNet, version 1.0.0, DOI 10.13026/C2F305, under the Open Data Commons Attribution License 1.0. The four records are 100, 105, 118, and 119. MLII is the primary ECG lead and the sampling rate is 360 Hz.

The public result preserves the established R-peak baseline and its deterministic one-to-one matching rule. The evaluation window is ±150 ms, equivalent to ±54 samples at 360 Hz. This package does not rerun the detector, benchmark, or notebook.

TP=9092, FP=28, FN=18. The error-stratification Build contains 46 event figures, 16 reciprocal counterpart pairs, and 32 linked events. Errors are concentrated in record 105 (44 of 46 events) and record 119 has 2 events. The canonical Build is 544c45cd1dbe33d85a5de8184c5594061a5fb605.

These values describe a bounded engineering pre-study on four public records. They do not establish clinical accuracy, diagnosis capability, medical-device performance, or generalization to the full database or a patient population.

本项目仅供教学、实验和辅助分析，不可替代医生诊断。