In the middle of the last century, long before the launch of the first artificial satellite, astronomers at the Palomar Observatory methodically scanned the night sky. The result was the grandiose archive of photographic plates known as the First Palomar Observatory Sky Survey (POSS-I), holding within it millions of points of starlight. For many decades, attentive researchers noticed something strange in these images: fleeting, starlike flashes that appeared and vanished within mere minutes. The science of that time often dismissed these anomalies as flawed emulsion, cosmic rays, or banal artifacts of chemical development. Yet modern technology compels us to look upon these "ghosts" in a completely different, more discerning way.
Double trouble -- our peer-reviewed machine learning/transient paper has also just been published. This is where we clean up the sample and see that the fewer contaminants there are, the stronger the correlation with nukes and the larger the shadow deficit. It also shows that the
A recent study conducted by an international group of scientists led by astrophysicist Beatriz Villarroel of the Nordic Institute for Theoretical Physics, and researcher Stephen Bruehl, has moved these observations out of the category of marginal curiosities and into the realm of rigorous science. Armed with machine-learning algorithms, the team set out to cleanse the historical data of decades of informational noise. A neural network, trained on thousands of pairs of images verified by human experts, analyzed more than 107 thousand previously identified transients. The model demonstrated high accuracy in distinguishing real objects from plate defects. The result proved astonishing: after filtering out the artifacts, the statistical significance of the anomalies did not disappear but was only multiplied many times over.
Two remarkable patterns emerged that are extremely difficult to explain by mere chance. The first was named the "shadow deficit." The number of recorded flashes declined reliably and significantly when the region of sky under study was in the Earth's shadow. This elegantly hints that the nature of these objects may be directly linked to illumination, or may be of artificial origin, somehow responding to sunlight.
The second pattern plunges the researcher into even deeper reflection, blurring the line between astronomy and history. The analysis showed that the probability of such transients appearing increases by 45% in the so-called "nuclear window" — within 24 hours before or after nuclear tests. The strongest correlation was observed precisely on the eve of, or in the first hours after, these fateful events of the mid-century. Moreover, the data revealed a weak but statistically significant connection between the frequency of these flashes and historical reports of unidentified anomalous phenomena (UAP) observations on those very same dates.
Particular significance is lent to these conclusions by the fact that on 12 October 2026, the final, fully typeset version of the work — the so-called Version of record — officially appeared in the authoritative journal Scientific Reports, part of the Nature Portfolio publishing group. In the academic world, it is precisely this moment that is considered the true birthday of a scientific discovery. Before that, the manuscript underwent rigorous blind peer review by independent experts, withstood all checks for methodological purity, and only then was cleared for publication. This means that the conclusions presented are not a bold hypothesis in the margins of an amateur forum, but a result recognized by the scientific community as worthy of attention and further study.
Today, looking at these digitized archives, we seem to catch the gaze of the past upon us. Perhaps these were atmospheric phenomena unknown to science, secret objects of the Cold War era, or something that still has no name in scientific classifiers. But the very fact of such fine synchronization with key moments of human history invites us to bold, prejudice-free inquiry. The universe keeps its secrets in old photographic plates, and now we finally have the tools to decipher them.
