Agreed, I've seen the same. Take that graph with a grain of salt and take a look at the failure modes table below it. Lots of monkey issues (about a quarter) and some totally unrelated things (like experiment termination).

Some labs/surgeons/methods have better success than others.

The big thing the sewing machine will do is (likely) reduce mechanical strain at the tissue-electrode interface and hopefully improve upon those handful of failures due to scar tissue encapsulation.

I had the same experience in my time at U. Pitt: arrays that worked at all worked for a couple years. I also spent a lot of time talking to people who had implanted Utah arrays at that time, and I found that there was a ton of variability in the details of how people performed the procedure. That was almost 10 years ago, so maybe there’s some stronger best practices now.