A second look at a 2026 Scientific Reports paper

Palomar's vanishing flashes

Old telescope photographs show tiny flashes that appear once and never again. A recent paper claimed they dodge Earth's shadow and bunch up around nuclear bomb tests. Rechecking the paper's own data, the numbers add up, but the conclusions fall apart.

A 1950s sky photograph, simulatedcircles mark "flashes"

The mystery

From 1949 to 1957, Palomar Observatory in California photographed the whole northern sky on glass photographic plates. On some of them are tiny star-like dots that show up in one photo and never again: not in the next photo of the same patch of sky, and not in any survey since. A computer search found 107,875 of these dots on 635 plates.

Nobody doubts the dots are there. The question is what they are: specks, scratches and flaws in 70-year-old photographic plates, or brief flashes of real light from something in the sky.

The 2026 paper, by Stephen Bruehl and colleagues, trained a computer model on 250 dots that an expert had sorted by eye into "real" or "flaw", and used it to score how likely each of the 107,875 dots is to be real. Then the authors reported three things:

  1. 1Fewer flashes where Earth's shadow falls in space. Sunlight can't reach inside the shadow, so if the flashes were sunlight glinting off objects in orbit, the shadow should have fewer. They reported 55% fewer among the most "real-looking" dots.
  2. 2More flashes within a day of U.S. nuclear bomb tests.
  3. 3Both patterns strongest for the dots the computer rated most likely real.

If true, that would mean shiny objects were orbiting Earth before the first satellite, Sputnik, was launched in 1957, and that they showed up around bomb tests. The paper's data turned out to be freely downloadable, so every step could be checked.

Claim 1 · does not hold

Earth's shadow almost never crossed a photo

Earth's shadow in space is a patch of sky about 36 Moon-widths across, always opposite the Sun, and it slowly moves across the sky through the year. The telescope rarely pointed there. In seven years the shadow overlapped only 11 of the 635 photos, so the whole shadow claim rests on 11 pictures.

Every photo in the study, and where Earth's shadow was that night
Night 1953-01-09 Photos touched by the shadow 11 of 635
One photo Taken that night Touched by Earth's shadow Earth's shadow Milky Way
A flat map of the whole sky, like a world map. Each square is one photo; the dashed oval is Earth's shadow on the night shown. Press play to watch seven years go by. When paused, the map shows 9 January 1953: that night one photo sat completely inside the shadow and still had 122 flashes on it.

Two mistakes made the shadow look empty. First, to work out how many flashes the shadow "should" have, the paper counted every photo as equal, even though some photos have one flash and others have thousands. Counting properly removes most of the gap. Second, the paper used a simplified shadow. Seen from California, the real shadow sits about 15 Moon-widths away from where the paper drew it, so the two only half overlap. With the real shadow, there are slightly more flashes inside it than expected, not fewer.

As a final check, fake shadows were placed at other spots along the shadow's yearly path. Using the paper's method on its most "real-looking" flashes, about a third of these fake shadows also come out at least 30% empty. With only 11 photos in play, an empty-looking patch of sky happens by chance.

Claim 2 · explained by the Milky Way

Spring skies point away from the Milky Way, and spring was bomb-testing season

To count as a flash, a dot must have no known star right next to it. Inside the Milky Way, the band of dense stars across the sky, there's almost always a star nearby, so most dots get thrown out. Photos of the Milky Way keep about 33 flashes each; photos pointed away from it keep about 200.

Flashes counted per photo
Typical count, by how far the photo pointed from the Milky Way (in degrees; 0° is right inside it)
Show numbers
How far from the Milky Way the telescope pointed
Average for each month of the year, in degrees; February to May highlighted
Show numbers

In spring, the night sky over California faces away from the Milky Way, so spring photos naturally count more flashes. The U.S. also did most of its 1950s bomb testing in spring: 47 of the 52 photos taken within a day of a test were from February to May.

So "more flashes near bomb tests" is mostly "more flashes in spring". Taking into account where each photo pointed and how it was taken, test nights go from 40% more flashes to about 10% more, small enough to be chance. And moving the test dates to the same weeks of other years, on days with no tests, gives a bump just as big about 3 times in 10.

Claim 3 · one photo

A single photo made the computer's "real" flashes look special

The paper's strongest number: among the flashes the computer rated most likely real, 13.6% were seen within a day of a test, compared with about 8% of the least likely. But one photo, taken on 3 May 1954, holds 799 flashes, and the computer rated 494 of them highly.

Share of flashes seen within a day of a U.S. bomb test
Flashes split into 10 equal groups by the computer's rating, from least to most likely real
All photos (as in the paper) Leaving out the one photo from 3 May 1954
Show numbers

Leave out that one photo and the most "real-looking" group drops to 9.5%, lower than most of the other groups. The trend was one night, not a pattern. And that photo only counts as "near a test" if a test in the Pacific is dated by Greenwich (London) time instead of local time. Use local dates and the trend disappears.

What's left