Is there a way for finer threshold control on the fill tool than what I'm seeing? Even the lowest setting still bleeds over shades that I didn't select.
It's a little frustrating, because compared to what I had used previously Krita is better in every way so far and I'm loving it, but this is a huge problem that makes it unusable for what I'm currently working on.

If there isn't, make this a suggestion - though I'd be pretty surprised if that were actually the case. Right now I just figure there's a trick to it that I must be missing.

Can you share a screenshot copy of the problem, with the fill tool activated and the tool option docker visible ? It might help to try to reproduce your issue.
Here I have no issue to limit the fill tool to a strict single color with Threshold to 1 and other option turned off...

Sure.
This may be a bit difficult to actually see on some monitors, but here's a screenshot from a little test to figure out a bit more specifically where the bottom end of the threshold was.

https://www.dropbox.com/s/rzqlmf3ccnn46 ... .png?raw=1

I filled in the top left box with Black (Exactly #000000), with the bottom half untouched and the markers to make it easier to see where the separating lines are on the bottom where the shades were closest. The lines are each an all around +1 increase in all three colours.

So, looking closely, even at the lowest possible threshold it floods over three of the lines before it sees the difference as large enough to stop it.
I did do a couple of extra tests too, noticing that interestingly - increasing only green, for example - it only jumped over an increase of 1.
So is this only an issue when working in purely grayscale colours? The green test maybe does fit the threshold value, though it's a shame it can't be lowered to 0 to effect only an exact colour, if that's the case. I may at least be able to use that knowledge to set up a workaround.

Thank you for taking the time to reply.

Hi, thank you for your experiment.
I could reproduce something similar, and so, confirm the bug.
I reported it here: https://bugs.kde.org/show_bug.cgi?id=385160