Ok, so what the tuner is telling you is that chain 1 and 2 were tuned successfully but then chain 0 had an issue with at least 1 chip, and thus it's trying to reduce the frequency of that hash board to get to a stable point, that also perfectly matches what wndsnb mentioned about one of your hash board showing a temp sensor error in the kernel log you posted earlier.

Eventually, the tuner might get it to a stable stage, but that could take very long and thus you are better off manually setting the frequency, use 600MHz on the two good boards and start with the lowest frequency on the bad board, and then slowly increase it until it becomes stable, keep in mind that sometimes even with the lowest frequency the hash board won't hash well which means you will need to unplug so that it does not affect the other two hash boards.

On a side but related note, the two working boards seem to do very well with a frequency of 600MHz, which means you could probably increase it, that gives you more hashrate and of course more heat, so you need to find that perfect spot.