Ab Astris: Cross-Domain Transfer of a Physics-Constrained Signal Detection Pipeline

This record contains the preprint of Ab Astris: Cross-Domain Transfer of a Physics-Constrained Signal Detection Pipeline (Tom Brady, The Modal Foundation, 2026). The pipeline source code is available at https://github.com/exs-brady/ab-astris. We present Ab Astris, a signal detection pipeline built around a fixed core: multi-window stability validation, bootstrap error estimation, and composite confidence scoring; with a default Lomb-Scargle spectral front-end that can be replaced where signal morphology demands it. Originally developed for variable star discovery using NASA TESS photometry, the pipeline transfers to five additional domains (bearing fault detection, volcanic tremor monitoring, oceanographic tidal analysis, wind turbine diagnostics, and structural health monitoring) with domain adaptation confined to preprocessing parameters and, in one case (planetary transits), a single-stage spectral substitution. Across domains spanning seven orders of magnitude in frequency, the coefficient of variation (CV) of detected frequencies provides a consistent empirical measure of signal reliability: mechanical systems cluster below 0.03%, tidal oscillation near 0.17%, structural resonance at 0.4–1%, and volcanic tremor at 1–27% (snapshot-level range across four eruptions). On the CWRU bearing benchmark, the pipeline detects 6 of 8 fault conditions with 0.21% mean frequency error, missing two weak or masked signatures that require domain-specific envelope analysis. On the Z24 bridge progressive damage benchmark (3,678 measurements, 23 damage phases), it tracks frequency degradation of up to 88% while maintaining zero false alarms on healthy baselines. Across six NOAA tide gauge stations, all 24 constituent detections achieve confirmed status with 0.005% mean frequency error. In astronomy, 2,947 periodic discoveries from a systematic survey of 60,638 TESS targets include 556 with mutual detections in the independent Fetherolf TESS Stellar Variability Catalog [Fetherolf et al., 2023], of which 517 (93.0%) show direct or harmonic period agreement (mean fractional difference 0.097%); 5 candidates have been accepted by the AAVSO VSX catalogue. Combined cross-matching against Gaia DR3, ASAS-SN, ZTF, and ATLAS confirms 1,563 (53.0%) of discoveries in at least one external catalogue, with 1,384 (47.0%) absent from all catalogues checked, of which 107 have no prior variability record in any of 20 variability databases surveyed. Three negative control domains (cryptocurrency, heart rate variability, sunspot numbers) confirm the CV metric’s discriminative capacity, producing values of 2.97–72% that are consistently separated from the physics-constrained band below 1%. The pipeline is positioned as a first-pass screening tool: broad, interpretable detection across domains, with domain-specific methods reserved where generic sensitivity is insufficient.