Full-census annotation-parsing risk audit across 16 PhysioNet ECG and interbeat-interval databases

A systematic, pre-declared full census of annotation-parsing risk across every database in PhysioNet's "ECG Databases" and "Interbeat (RR) Interval Databases" categories (16 databases, no exclusions), extending the single-database finding reported in an earlier full-corpus audit of the STAFF III and LTST databases (https://doi.org/10.5281/zenodo.21501981). Four annotation-parsing risk classes were defined and tested: missing-token assumptions (R1), null-byte padding of WFDB annotation strings (R1b), case-sensitivity assumptions (R2), and symbol or field reuse across annotation files (R3); a fifth class, free-text clinical-field inconsistency (R4), was tested in one database. Ten of the 16 databases contained a testable annotation family; seven showed at least one confirmed, reproducible risk instance, including one database (MIT-BIH Malignant Ventricular Arrhythmia) in which 100% of rhythm-annotation instances were affected by null-byte padding, and another (MIT-BIH Arrhythmia Database) in which 95.7% were affected. Three databases showed no risk under any criterion tested. Six databases were structurally not applicable, containing only beat-level detection output. Each database has its own folder containing a file-discovery inventory, a structural applicability determination, a risk-scan result table, and the underlying raw evidence supporting every scored result, so that all reported numbers can be independently recomputed. A top-level CENSUS_SUMMARY.csv aggregates one row per database. The exact code used to build the census (record inventory retrieval, WFDB annotation reading, risk computation) is included. Only header and annotation files were retrieved from PhysioNet for every database; no raw ECG signal files were retrieved or required, since all tested risk classes are properties of the annotation stream rather than the underlying signal. This dataset supports a companion Data Descriptor manuscript submitted to Scientific Data.