Work on better calibration. Make sure to not term "impossible" what will turn out to be true. Be aware of the inferential distances: maybe you just don't get the intended meaning (but this is also a problem with presentation).