The problem was a weight calculation error. This still is not great though. My adaptive sampling code doesnt exactly work the way I want yet with the weighted samples but even if it did it would look kind of like this image, which is just 64 samples per pixel. I have been researching what type of sampling to do with reconstruction filters and perhaps halton is not the best. The radius of the box filter is 2 pixels wide, and this seems to blur the image too much even though values more than one pixel away are being subtracted. You cant even really see the sharpening anywhere in this image.

Changed up my algorithms and box filter range but now I get some interesting results. Gonna set this aside for now I think

Tried adjusting the delta value based on the weight of the samples I took and it still doesnt look great. Gonna try reducing the box that contains the catmull weighting to 1.5 radius instead of 2 Update, that didnt fix it. Im not entirely sure what to try. I think that a post process effect may really just be better, or collecting all the samples in a vector rather than trying to come up with my own math for an alternative to Wests algorithm that works with weights. Update again: just found the correct version of Wests algorithm that works for weighted sums

First attempt at reconstruction filter with adaptive sampling that works kinda. I think I know what I need to do to fix it.

Sample count image for previous result

1080p doubling samples, 8-64. 4 second render with 96% confidence interval which contains many less errors than 95%. 98% takes about 8 seconds.

1080p render with range of 8-64 samples, takes 3 seconds doubling the number of samples when below threshold and 4 seconds with an additive technique adding 8 samples when below threshold.

Sample count image for the previous result. Range 8-64

Adaptive sampling doubling number of samples when past threshold. 95% confidence with 0.01 delta max and min number samples of 8 and max of 64. 2 seconds to render.

64 samples per pixel, 7 second render macbook M1 pro