I understand that salmon does not use read quality scores when performing alignment/quantitation but, as it stands, running with --writeMappings it also doesn't output the qualities in its SAM-format output.

That may not impact most downstream applications of the data but it does limit some. I'd be particularly interested in using reads that have been aligned and assigned to transcripts with an EM algorithm to investigate RNA editing. The most advanced methods of doing this integrate the quality of base calls from the sequencer into the models and, as such, cannot currently run on the data output.

Although it may save a few CPU cycles in not reading and outputting the qualities - I can understand the choice when running with default settings - when running with --writeMappings that nominal difference in speed needs to be weighed up against downstream utility. Could this be looked into?

Cheers! George