Here a composite of 66909 and 66908 Doppler curves compared against SilverSat CW beacon transmissions from the most recent pass over the USA. Note the near perfect correlation between 66909 and SilverSat–it is surely object 66909.
As far as one TLE producing “better observations”, this may be due to many factors (Fredy has already mentioned.) These factors (not exhaustive) include both those at the satellite (eg. it’s antenna pattern leading to varying signal strength as the satellite’s orientation changes with respect to the observer) as well as the ground station (eg. antenna polarity, beamwidth, pointing accuracy/step size, nearby objects, ground gain, and the inputs that drive the antenna system.)
To predict the satellite’s position accurately the model inputs used must be the most recently derived and the observer variables must be too! The observer’s station location and clock must be exact to obtain the best result.
Doug K8DP