The view through the front window (1950-vintage glass + storm window) at 1/500 s, f/4, ISO 100, auto-exposed to EV 0:
The same view, using stacked 2 + 4 + 8 ND filters, at 1/3 s, f/4, ISO 100, auto-exposed to EV 0:
That blur to the left of the mailbox is a passing car, which is the whole point of heavy ND filtering: you can use absurdly long shutter times and still get a decent exposure.
I cannot explain the fact that the ND filters allegedly reduce the light by 14 stops, but the actual (auto) exposure increases by about 7 stops. They’re cheap non-coated K&F Concept Digital filters, of course, but …
There’s an obvious color shift toward red / magenta.
This is a placeholder so I can pick up the thought later on…
The Smell of Molten Projects in the Morning 
I found the mail box but have a problem with the blurred car. Is there something I’m missing?
That’s how blurred it is!
Figure 40 mph = 60 ft/s, so a 1/3 s shutter lets the car travel 20 ft. There’s a blue streak that starts just above the mailbox and extends leftward to about halfway between the utility pole and the tree; the car was traveling right-to-left in the far lane.
Your snow looks terrible!
And it’s getting shabbier, even if the packed pile out back is still a foot deep…
hypothesis: your ND filter passes IR, which is detected primarily by the red colorsites on the bayer sensor. A more expensive ND filter or an IR-block filter in the filter stack might improve things. see e.g., http://www.tiffen.com/userimages/IRND-FullSpectrum_SS.pdf
[Ed: Added linkification.]
That’s the old DSC-F717 with a Nightshot IR mode, so it may be hypersensitive to red even with the internal hot mirror in place. Sony engineers looove saturated red, as you’ve seen in all the pictures of red-filament 3D printed stuff, and this may be more of the same.
Perhaps the stash has some IR block filter material…
I just reviewed the reciprocity-failure article in Wiki, and recall seeing examples of severe color shifts in low light. Looks like charge bleedoff is the factor in film, and while the article is silient on this, I suspect the wavelength/bleedoff curve isn’t dead flat. No reason to assume a solid state sensor would be immune, though a really cold sensor should be less sensitive to failure effects. I think.
I’d also wonder just how neutral the ND filters are. If you have a 60 degree prism on hand, I’d cobble up an optical bench setup with a wideband photodiode mounted on the Sherline stage so you could pick up the spectrum of the light over an X axis. Place the ND filters in the path and see what happens. I think you could keep the intensity high enough to avoid reciprocity failure effects in the PD… [sigh]
Or maybe I can put the digital spectroscope tinkering to good use. Hmmm… sounds like a good Circuit Cellar column idea: thanks!