Thanks! This was enough to point me in the right direction. The nonlinear intensity encoding of sRGB was the key. Blender’s own color correction on renders (seems like Filmic by default on 2.82) also contributed a little bit.
Turns out the bright parts of the chest in the first picture above have a raw intensity of around .18, which is encoded as .47 via sRGB. But that’s with Filmic color management, with standard the resulting image’s bright have respective intensities of .14 raw / .41 sRGB.
Is it okay to multiply the grayscale colors by a constant via a math node before passing into the colorramp node, just to make the colorramp easier to manage?
