Great rant, thank you. I agree with most of the problems you're outlining.

First, on layers - I never suggested that we only need one single set of trust assignments. If we need several layers or facets - we could have more.

Second, on updating - manual or automated, any trust signal that is published will have to be updated periodically. UX might make it super smooth, but still. 

Third, on discrete flags - instead of 0-100 scale you're suggesting 0-1 scale, what does that change in principle? If humans are limited - let them only publish 0 and 100, and leave the full range to machines. 

Is there a preview of the NIP you're working on?