I was foolish and deluded. The wipe results in all the light redder than the pixel being included. I used the mean for the wipe model, and in conjunction with the all light bluer from the read the model was flexible enough to work, but with the wrong times for the coefficients.

Fix the code to use the proper model, and update the coefficients to be physical interpretable.

While we're about it, set the variance correctly when using the analytical correction; and don't try to correct biases.