The universe's expansion rate, measured by objects you can hear but can't see.
Dark sirens: gravitational wave events without electromagnetic counterparts. Unlike bright sirens (like the neutron star merger GW170817, which produced light and directly measured both distance and redshift), dark sirens provide distance from the gravitational wave signal but no direct redshift. The redshift must be inferred statistically from galaxy catalogs — which galaxies could plausibly host the source?
Note (2026-11-15; read against the paper's full text): The 11 percent figure is not what the seventeen dark sirens achieve on their own. In the paper it is the improvement in the uncertainty when the luminosity-weighted dark siren sample is combined with GW170817 (with jet viewing angle and peculiar-velocity corrections), relative to GW170817 alone; the dark sirens alone give 78.2 with plus 12.0 and minus 11.0, a much wider range. The paper also reports that individual events differ: one event (GW231206_233901) is the most informative, with a 1 sigma constraint reaching about 77 percent of the prior width in the unweighted case, while others are weak, so 'individually contribute almost nothing' is the essay's own summary.
The new measurement (arXiv:2603.20195) uses 17 well-localized dark sirens from LIGO-Virgo-KAGRA runs, combined with the DESI Legacy Imaging Survey and deep-learning photometric redshifts. Result: H₀ = 78.2⁺¹²·⁰₋₁₁.₀ km/s/Mpc from dark sirens alone. Combined with the bright siren GW170817: H₀ = 69.9⁺⁴·¹₋₄.₀ km/s/Mpc — an 11% improvement over GW170817 alone.
The methodological advance: luminosity weighting (brighter galaxies are more likely to host merging black holes), binary mass information (the black hole masses constrain the source population, narrowing the redshift inference), and catalog incompleteness corrections (accounting for galaxies too faint to appear in the survey).
The structural point: dark sirens individually contribute almost nothing — each one's redshift uncertainty spans a huge range. Their power is statistical. No single dark siren constrains H₀ usefully. Seventeen together narrow the measurement by 11%. The information is in the aggregate, not in any individual event. The census works where the portrait doesn't.