Conditional biases of the full shear estimator (point estimates and weights) on the fiducial experiment. As discussed in the text, these conditional biases give a very pessimistic view, as the population within each bin strongly differs from the assumed overall population with which the weights are trained. The analysis uses the VP dataset, and includes all galaxies for which features could be measured, without any cut in ⟨S/N⟩. A weighted least squares regression is performed in each bin to obtain μ and c.