Ranger: I think you might be right that there is still a bit of an edge artifact there, although it’s quite subtle compared to the one on top. I’m a little surprised by this, but I suspect it’s a little sharpening built into the Bayer-pattern deconvolution of the pixel data. As I’m sure you know, digital camera sensors (and I’m leaving out Foveon sensors here) only pick up a red, green or blue value at each location, the other primaries for each location are computed using an interpolation function, essentially by smoothing, which tends to blur the chroma values of the resulting image, it would not surprise me if a pinch of sharpening was implicit or explicit in this part of the calculation in Lightroom.

While I agree with you that the yanking around of pixel values quote-unnaturally-unquote is part of what’s triggering the stairstep artifact, it is my guess that there is arguably a bug there as well. Lightroom has an incredibly difficult job in image processing, Bayer-pattern deconvolution, sharpening, tonal adjustments/grayscale adjustments and noise reduction all have to be performed on some images and as with any sort of digital image processing, the correct order of operations is essential for best results. If noise reduction were the last step, it’d be pretty much ridiculous to look at the “no noise reduction” and to suggest that “noise reduction” would produce the artifacts shown above.

While it might be that a different internal order of operations would resolve this, I think from other experiments that the problem is actually a bug, for a variety of reasons (different results for situatons which are more or less mirror images of each other, size of stairsteps being more than one pixel in any dimension, etc.)

(Of course, I don’t know of another raw converter that offers all this functionality to demonstrate the differences on. So it goes!)

This is happening because photographic images generally consist of statistically distributed ranges of pixel values, rather than mathematically ideal regions of constant or smoothly changing pixel values. In an idealized, smoothly gradated image (such as one you might create with illustration software) pixels which are close together will have similar values. In real-life photography, any two pixels that are arbitrarily close together may have drastically different values.

When you apply mathematical transformations to a specific range of pixel values without affecting others — which is what the grayscale-mix adjustments do — you can easily shift those values far enough to create an unnatural-looking relationship to the values of pixels near them. (This is why Photoshop retouchers have to use the “Add Noise” filter to put variation back into areas on which they’ve painted heavily.)

The practice of applying transformations to some pixel values but not others is also the underlying cause of the other two problems you mention, as well as many other artifacts that crop up in image manipulation. The more selective you try to be about altering pixel values in your images, the greater the risk that the altered pixels will no longer have a realistic relationship to others nearby.