Suggestion: remove the health deduction from Swahili on the mobile app until the problems are fixed
I just completed a one-lesson unit in Swahili, and there were so many problems with what translations were considered acceptable that the only way to finish the lesson was to memorize the bad translations that were accepted. If I were doing this lesson on my phone, I would have run out of health long before I got anywhere near the end. This problem has cropped up for a few lessons, and I think it would be best not to use the health counter on the mobile app for Swahili--at least until the tree has been updated with more variety in the accepted translations.
Maybe take this up in the troubleshooting forum rather than here? I don't think those with the power to do that check the individual language forums.
Not that I think they would, but maybe it's worth a try. If you do, maybe bring up that Swahili team thought Duo was adding the audio as well.
Like juryrigging, I think this post is very unlikely to reach its intended audience here.
You don't have to convince anyone familiar with the Swahili tree that it's strange it made it out of beta.
I would think this actually involves some sort of change to site features (for which the default forum is the main, overall one for English speakers). Evidently there's something about the current system for releasing courses from beta that is letting courses that pretty obviously aren't ready slip through the cracks.
Maybe error report rates for individual skills throughout the tree should be considered instead of just overall rates. For a tree like Swahili, I suspect many users don't get all that far, so they mostly do the skills that are in (relatively) much better shape. This dynamic exists for all trees, but the particular difficulties of the Swahili tree could easily exacerbate it.