Open Veils · Answer
Any specific anomalous spectroscopy data flagged?
Multiple independent machine-learning pipelines applied to DESI, SDSS, and MaNGA datasets are now surfacing a residual population of spectra that survive artifact filtering and remain genuinely unexplained. The scale is not trivial: approximately 218,000 candidate outliers in DESI DR2 alone cannot be explained by known reduction effects. This is the kind of dataset that researchers like Garry Nolan and Jacques Vallée have argued mainstream science is structurally unprepared to interrogate — anomalies that don't fit a category simply get deprioritized.
Evidence
- A June 2026 arXiv preprint applying UMAP and Friends-of-Friends clustering to 58 million DESI DR2 spectra found that standard pipeline quality flags caught only 4.1 percent of anomalous spectra that the unsupervised method flagged — meaning the conventional toolchain misses the overwhelming majority of outliers.
- Extrapolating across the main-survey catalog, the authors estimate roughly 218,000 spectra free of identifiable reduction artifacts, representing genuinely atypical astrophysical signatures.
- Separately, a University of Portsmouth team using a Variational Autoencoder on ~200,000 DESI galaxy spectra corroborated anomaly clustering that spans galaxies, quasars, and stars simultaneously.
- A November 2025 Astronomy & Astrophysics study applying similar methods to SDSS's 791,000+ spectra found a comparable residual population.
- An April 2026 Publications of the Astronomical Society of Japan paper on the MaNGA Stellar Library flagged carbon stars and thermally pulsating AGB stars as reconstruction-error outliers — known exotic classes — but acknowledged the method's limitations in distinguishing instrumental from astrophysical sources.