adrianklein first of all, thank you very much for opening new topic for new issue. This is not very common and I typically have to split topics manually all the time. I really appreciate that.
Regarding your low confidence idea, unfortunately, this is way more complicated that it may seem on the first glance. The thing is, there is not so much correlation between confidence and what is correct suggestion and what is not. If it would correlate as good as it needed for your suggestions to work good, I would just rely on confidence only and would not force people to go thru all the pain with accepting suggestions, I would just automatically accept high confidence results. But unfortunately, it is not working that well, I tried that earlier, but in the end I had to completely disable this auto-accepting feature, since from time to time it did strange things, that I was not really able to explain to users. Also, apart from just face features, I started using some other context-based features too, which helps me to improve suggestions quality.
I get requests like that from time to time, and I assume some time in the future I will end up doing something similar to what you are suggesting that would help “debug” already accepted faces. However, that is not going to be simple. And the challenge is, I have to make it the way it would be easy to use for most of the users. I definitely do not want to turn Tonfotos into scientific tool for nerds. With all that said, difficulty of this feature is hight, but most users would never bother to be so meticulous about their library, that is why this suggestions is not among high priority ones, unfortuntaltey.