So after letting the last version sit for a day or two, I started having doubts about the values. It just didn't look right; overall too light in the mid-tones.
This is a good time for the aid of modern technology. I made a grey-scale version of the last progress photo I took (on the left), and compared it to my original value study (right):
Of course there were some differences in the lighting when I took the shot, and that sort of thing, but I can definitely see that my recent version is too light and without contrast.
Using this as a rough guideline, I darkened some areas and lightened others:
This is much closer; I think I can move on from here. My original study only used five shades, whereas I'm not limiting myself to that in the final painting. Still, I feel like this is more on track — darker overall, and more contrast between darkest and lightest values. There are a few questionable areas still, but I think I can compensate in the next iteration. I think that one light piling just-right of center is important to the composition; it's one of the brightest spots in the lower half, and it's sort of pointing you in to the areas of interest — it leads you up the stairs into that very dark doorway.
The great thing about working out the values ahead of time, is that when I go to paint the real colors, I can easily see if they are too dark or too light. If I put a dab of paint on the canvas, I can squint my eyes and tell if it blends in. If it does, the value is correct; if it doesn't, I need to try again.
That's why, if you get value correct, the color doesn't matter: you can make the sky green and the grass pink, if their relative values are right.