Double trouble -- our peer-reviewed machine learning/transient paper has also just been published.

Beatriz Villarroel shared a peer-reviewed study using machine learning to distinguish short-lived, star-like features in historical photographic plates from plate defects. The authors report an association between these candidate astronomical transients and nuclear-test dates, alongside a deficit in Earth’s shadow. The strongest timing association occurs on the calendar day before a test—not necessarily 24 hours earlier: the paper cautions that overnight exposures may precede a morning test by only several hours. These findings do not establish that the transients were alien craft or explain their origin, and the authors call for independent replication. Social-media comparisons with the 2004 Nimitz Tic Tac encounter are interpretations, not findings of this study.