BiSON Sun-as-a-Star Data: New Calibration Method for Long-baseline High-cadence Extreme Precision Radial Velocities

For over four decades now, the Birmingham Solar Oscillation Network (BiSON) has provided Sun-as-a-star radial velocities with 40-second cadence and near continuous coverage, becoming one of the cornerstones of helioseismology. Yet, this extraordinary dataset has never been explored for solar effects with timescales longer than one day: standard calibration procedures intentionally suppress low-frequency signals, erasing information about solar variability on convective, rotational, and magnetic-cycle timescales.We now present a complete re-calibration and re-analysis of the full BiSON archive designed to retain these ignored signals while preserving BiSON’s unmatched precision. We introduce a generative spectral model coupled to a physics-aware neural network framework for multi-decade, multi-instrument calibration. This approach isolates local instrumental and terrestrial systematics while enforcing a single, coherent solar activity signal across the global network, enabling robust separation of solar variability from non-solar signals and noise without imposing restrictive parametric assumptions.The resulting product is the longest-baseline solar radial-velocity time series to date with short cadence high-precision data. It opens new avenues for detecting elusive low-frequency seismic signatures, including g-modes, and provides an unprecedented laboratory for characterising granulation, supergranulation, and magnetic activity across multiple solar cycles. As such, this dataset also offers a critical benchmark for developing stellar activity mitigation strategies central to the detection of habitable rocky worlds with extreme-precision radial velocities.