8-128 samples, confidence 98%, max difference .05. Note, theres some glitch at the bottom apparent here, due to some changes I made with ordering ray calculations. Also, this is with the project 9+10 code.

finally with adaptive sampling

Changed random distribution to get random vector lerped between the top and right pixel vectors vs. the current one. Previously, was using a dot product converted into degrees. This is at x16 sampling.

x8 sampling, clearly not functioning correctly

4 samples, full random offset