Purpose: Continuous high-density physiological data from bedside monitors are rarely captured systematically. We implemented and audited a large-scale acquisition system in a real-world surgical intensive care unit (ICU), characterized its failure modes, and compared the dataset with international benchmarks.Materials and methods: Single-center descriptive implementation study in a 13-bed surgical ICU. A decentralized, clinician-led architecture using open-source software (VitalRecorder) on existing bedside computers captured signals at up to 500 Hz from Philips IntelliVue monitors during a pilot phase (May–December 2024) and a 12-month prospective period (January 2025–January 2026). An automated pipeline audited coverage, signal availability, quality, artifacts, fragmentation, and failures. Monitoring phenotypes were compared with the MIMIC-IV Waveform Database.Results: 5,889 files were collected (23,203 h of monitoring; estimated 250–300 patient episodes). ECG and SpO₂ were available in 94% of files with artifact rates ≤ 0.01%; invasive arterial pressure in 48.4%, capnography in 22.7%, intracranial pressure in 20.1%, and EEG/bispectral index (BIS) in 6.9%. Median completeness was 95.7% (IQR 83.1–99.5). Capture varied across beds from 7.9% to 61.1% of eligible bed-days, and nine recurrent failure modes were identified. Signal co-occurrence differed from MIMIC-IV (Jaccard distance 0.85), monitoring diversity was higher (Cliff's delta −0.52, p < 0.001), and five monitoring phenotypes differed between datasets (χ² = 115.1, p < 0.001).Conclusions: Large-scale high-density acquisition in the ICU is feasible with existing infrastructure and open-source tools; the main barriers are organizational rather than technical. Differences from public waveform resources highlight the need to characterize local monitoring practice before developing signal-based models.