FAST's signal pipeline keeps the candidates and the vetoes visible

Researchers analyzing FAST observations of 33 exoplanet systems built a pipeline around a multi-scale wavelet network, endpoint regression, morphology filtering, raw-data signal-to-noise checks, and a multibeam anticoincidence veto. Instead of relying only on a hard drift-rate threshold, the workflow represents narrowband features at multiple scales and preserves interpretable steps that analysts can audit.

Applied to real observations, the system recovered representative events from earlier analyses and produced a smaller set of candidates ready for additional vetoes and human inspection. The paper does not report an extraterrestrial signal. A candidate that survives one automated stage is a research queue entry, not a discovery, and recurrence plus independent confirmation would still be required.

Why it matters

Large radio surveys need automation, but a transparent chain of thresholds, vetoes, and raw-data checks makes it possible to understand why a signal survived—and why it may later fail.

What the source establishes

Worth Revisiting · Peer-reviewed methods paper in The Astronomical Journal. The pipeline and reported validation are documented; no technosignature detection is claimed.