I suppose the prediction didn't include a scale (it was determined from the observation+model) so there is a little bit of room for model fitting there. And sure, other stuff could have generated a frequency sweep. Maybe if this is replicated the newer methods will narrow down the possibilities.