Yeah, we can’t a priori assume that a divergent sequence is wrong, but for all sequences in this set where there were multiple mutations, many (all) of them were clearly the result of sequencing errors. Even after taking out the most ‘offensive’ sequences, I’m still fairly certain ~50% of SNPs are sequencing errors.
Doing some quick calculations on a SARS alignment from the 2002 epidemic, I get an N/S rate of ~0.5 and a Ts/Tv of ~ 3.5 - meaning that while synonymous and/or transitions are more frequent, but maybe not as much as we would typically see with other RNA viruses (e.g., Ts/Tv typically ~8). Just as a reference for trying to weed out sequencing errors and bad sequences.