Predictions based on TLEs aren't precise enough for satellite collisions. TLE is a solution compiled from many measurements, and what one wants for very close p

Predictions based on TLEs aren't precise enough for satellite collisions. TLE is a solution compiled from many measurements, and what one wants for very close passes are raw measurements. It's a bit like JPEG vs RAW in photography. Regarding that particular collision, IIRC human factor was involved a lot. Iridium has been warned well in advance, but hesitated to make the decision since their models gave lower probabilities.