Next-generation cosmology faces a paradox: the demand for sub-percent precision clashes with unsustainable computational costs. As BINGO approaches first light and SKA construction advances, reliance on GPU-intensive machine learning for RFI rejection threatens both carbon budgets and reproducibility. We propose a paradigm shift: a physics-driven, content-agnostic transient rejection framework based on a single fundamental quantity — the Cosmic Rhythm (τdwell), the natural timescale imposed by Earth's rotation. Unlike black-box classifiers, our filter uses first principles (τdwell = θFWHM / vdrift) to distinguish celestial signals from anthropogenic transients. The framework was validated via Monte Carlo simulations (10⁶ events), BINGO HEALPix mocks, HalfDome Stage-IV lightcones, and the First CHIME/FRB Catalog (CHIME/FRB Collaboration 2021), achieving 100% rejection efficiency for real astrophysical transients in the BINGO band (980–1260 MHz). Overall rejection efficiency exceeds 99% for impulsive contaminants with negligible false positives on cosmological signals. The algorithm is deterministic and scales as O (N), eliminating GPU dependency and aligning with Green Computing standards. An integrated Instrumental Event Provenance (IEP) module ensures full auditability and scientific recovery of borderline events via SHA-256 hashing. Wavelength-agnostic, the method generalises from radio drift-scans (BINGO, SKA) to optical surveys (LSST, Simons Observatory). This repository includes the preprint (PDF) and the validation notebook (Jupyter/Python). Notebook v1. 2. 0 (2026-03-21): Algorithm 1 refined — true running median via scipy. ndimage. medianfilter replaces mean approximation; local MAD estimator introduced for non-stationary noise robustness. False alarm rate characterised (< 0. 05 events per 2 h BINGO simulation at 5σ) ; IEP snrₑstimate field annotated for noise-driven event identification. Preprint and scientific conclusions unchanged.
Alexandre de Sá Vieira (Sat,) studied this question.