The decorrelation afterburner makes the template and science image roles symmetric in their noise propagation in the AL image subtraction. Without this, the template is assumed to be noise free.

Considering that there are objections to deconvolution (theoretical limits of inversion, sampling criterion, spacial smoothness of the solution; assumptions of the Gaussian approximation paper) and it also performs poorly on the decam images in practice, we may want to update the automated way of image differencing.

We may want to use the doConvolveTemplate=False + doDecorrelation=True path automatically instead.