The most precise inference of the Hubble constant uses a three-rung distance ladder (geometry to Cepheids to supernovae), where the supernovae are needed to probe the Hubble flow where peculiar velocities are negligible. However, recent advances in the Bayesian Origin Reconstruction from Galaxies (BORG) algorithm have provided highly accurate local peculiar velocity fields with precisely-characterised uncertainties, enabling quality constraints on the Hubble constant without the third rung. I will describe a hierarchical Bayesian forward model to infer H0 from SH0ES Cepheids and geometric anchors alone, marginalising over the Cepheid period-luminosity relation, galaxy distances and various nuisance parameters of the velocity field, including a statistically rigorous accounting for selection effects. For the fiducial selection model the result is H0 = 71.7±1.3 km/s/Mpc, slightly lower than (though consistent with) the SH0ES result and discrepant with the CMB-inferred value at 3.3 sigma. Alternative selection models produce at most a 1-sigma shift. As well as supporting supernovae as accurate contributors to the Hubble tension and highlighting the vital importance of robust peculiar velocity fields, this result demonstrates great promise for future two-rung H0 inferences incorporating more data. I will also stress a few general statistical aspects of distance-ladder modelling, particularly the need for an r^2 prior on distances and a principled selection model — and how badly things can go wrong if inadequate statistics are used.