Take the BOSS galaxy samples, measure their clustering and stellar masses, fit the standard galaxy-halo connection under Planck ΛCDM, and the model tells you exactly how much gravitational lensing those galaxies must produce. The measured lensing is 20 to 40 percent lower. Leauthaud et al. (2017, arXiv:1611.08606) established this "lensing is low" discrepancy for the CMASS sample, and subsequent work extended it to LOWZ and across independent imaging surveys: the deficit is largest, around 30 percent, on small scales (r_p below 5 Mpc/h), and a blind comparison across five lensing surveys (Lensing Without Borders) confirmed the amplitude is not an artifact of any single dataset.
The escapes have been tried in order and found wanting. Leauthaud et al. showed baryonic feedback, massive neutrinos, and reasonable modifications of GR each move the prediction by only a few percent, far short of 30. More flexible halo-occupation modeling, including assembly bias and halo-mass-dependent selection, can close the gap on the smallest scales (below 1 Mpc/h) but strains to cover the full range, and the parameter freedom required begins to undermine the predictive content of the halo model itself. The residual discrepancy points the same direction as the S8 tension: the late-time universe lenses less than a Planck-normalized ΛCDM expects from its clustering, whether the tracer is cosmic shear, CMB lensing cross-correlations, or galaxy-galaxy lensing.
The standing is mature and quantitative: lensing-is-low is one of the best-cross-checked small-scale anomalies in cosmology, and full-shape BOSS reanalyses now treat the lensing amplitude deficit as a parameter to be explained rather than a bug to be fixed. DESI lensing cross-correlations and Euclid will measure the scale- and redshift-dependence of the deficit precisely, which is the discriminating information every proposed explanation must now match.