On top of your analysis of commands that might flag conflicting commands, shouldn't we also be analyzing the radar to make sure there aren't any obviously conflicting paths?
I saw your comment about pilot intentionality being the major time-saver but in a situation like the Potomac River crash, the intentionality and knowledge may have been a unintentional headfake for a system like yours i.e "Yes I see the CRJ".
Maybe it's 80-20 for you, with the Potomac situation being especially rare.
Your general point is 100% correct. It's not enough to only do the command parsing; you have to also compare that to radar + trajectory planning to get the full picture. That's exactly what we're working on now. Fortunately, the radar piece is much more deterministic.
On a clear day though, it's probably just people looking out the window of the tower (maybe with binoculars) and realizing in horror that there's been a mistake. The "noticing" is just intuiting the trajectory of the two moving objects...