This work investigates how acquisition context and photoplethysmography (PPG) each contribute to session‑calibrated blood pressure (BP) tracking. A cohort of 653 AuroraBP participants was evaluated with three repeats of five‑fold participant‑level cross‑validation, excluding calibration‑labelled prediction targets. Matched analyses separated context from PPG and removed elapsed‑time terms. Follow‑up explorations retained PPG, accelerometer signals and the first three cuff readings, while separate studies examined pulse‑rate fusion and an intra‑operative VitalDB model.
The original wearable‑state model achieved participant‑level mean absolute errors (MAE) of 7.697 mmHg for systolic and 6.139 mmHg for diastolic BP. Adding context reduced MAE by 0.436/0.126 mmHg, with 97.5 % intervals of 0.305‑0.573 and 0.053‑0.200 respectively. In the matched factorial without the external PPG prior, PPG improved over calibration, time and protocol‑provided posture/activity by 0.214/0.135 mmHg; removing time left the DBP interval including zero. Standard PPG descriptors contributed, but no additional nonlinear/attractor benefit was established.
A sensor‑history ensemble in a later series yielded MAE of 7.299/5.445 mmHg versus a static‑rich reference of 7.363/5.470 mmHg, a gain of 0.064/0.025 mmHg, with larger DBP errors not improving uniformly. Quality‑weighted fusion reduced pulse‑rate MAE from 3.985 to 3.690 bpm. The VitalDB model’s MAE remained 15.375/8.833 mmHg despite modest PPG gains.
The significance lies in providing a calibrated evaluation framework with explicit controls, reproducible feature‑cache‑to‑results analyses, and the preservation of negative findings. The sensor‑history gain is exploratory after repeated dataset reuse. Independent wearable‑state estimation, full raw‑signal reproduction, external transferability and clinical adequacy remain unestablished.
Review