The problem isn’t strictly with obscuring, it’s the combination of obscured and large accuracy values. With obscured observations, there’s an accuracy value given, but the circle itself isn’t visible (see https://forum.inaturalist.org/t/obscuring-rectangle-on-obs-with-large-accuracy-circles-may-not-encompass-true-location/703/19 for a suggestion on how to fix that). For observations that are both obscured and with large accuracy, it’s unclear whether it’s appropriate to mark not accurate in the DQA or to let them continue flagging the atlas. It’s possible the accuracy circle does include part of the atlas even if the “true coordinates” aren’t in it. (If that didn’t make sense, I can send a better description privately so as not to “give away” any tips.)
Related, I’m not sure I agree with how you’re using “true coordinates”. For something with a 1000 km accuracy circle, the true coordinates are the center, but that doesn’t guarantee the organism was seen anywhere near there.
At any rate, a 5-10 km cutoff should fix it for both obscured and unobscured observations.