Imagine a lottery where instead of money, you win black holes. Each ticket is a random spot on the sky, and the prize is a gravitational wave from the merger of two giants. The chance of hitting the jackpot is minuscule, but if you do, the wave is heard across the entire universe. That's how the gravitational-wave background works: it's the collective hum of many distant supermassive black hole mergers, picked up by networks of radio telescopes that monitor pulsars.

Until recently, astronomers estimated this hum using average values, as if all lottery tickets were equal. But reality is different: rare, bright sources create shot noise — random fluctuations that distort the background map.

The authors generated 700,000 random populations of merging black holes and computed the shot-noise amplitude for each. It turns out that the distribution of this amplitude is broad and asymmetric: typical values are 2–3 times below the mean, and at high frequencies the mean goes through the roof.

Why does this happen? It's all in the math: the mean of a ratio of two quantities is not equal to the ratio of their means. Simply put, if you calculate the average lottery payout, rare jackpots skew the average upward, even though most players win nothing. In our case, moment-based estimates (using averaged powers of the amplitude) overestimate shot noise by a factor of 3 at a frequency of 0.1 yr⁻¹ and by more than 100 times by 1 yr⁻¹. The median shot noise grows from 3×10⁻³ at 3 nHz to 0.05 at 30 nHz, following a power law of $f^1$.1–1.2 instead of the expected $f^8$/3.

This means that when planning observations, you can't rely on average estimates: the real sky map will almost certainly be “emptier” than the mean predicts. The scatter between different realizations is enormous: the 95% interval spans a factor of ~50, meaning two random universes with the same parameters can differ in shot noise by a factor of 50.

The work changes how we interpret pulsar timing array data. Previously, scientists compared measured anisotropy to the theoretical mean and drew conclusions about the black hole population. Now it's clear: you need to model the full distribution function of shot noise, or you might easily mistake a random fluctuation for a signal of new physics. This is especially important for separating contributions from astrophysical mergers and possible cosmological sources of gravitational waves.

The lottery continues, and each new ticket is a new realization of the sky we observe. So far, no one has hit the jackpot, but knowing that the odds are unevenly distributed makes the game fairer. Returning to the opening image: in the gravitational lottery, most tickets are blank, and only by accounting for this fact can you correctly read the sky map.