Not waiting on anything except my own time to move it forward.
The challenging part for me was that I felt like nobody really tried to understand what I propose and almost all the discussion was about something else than the gist of the original post. And for the only questions that I believed were on target, I gave answers (at Separation of concerns in autodiff tests - #8 by martinmodrak) and then got no further feedback while people continued to respond to what I believe were minute details raised by others. It felt as if my thoughts are invisible. It is well possible I didn’t explain myself well, but it would be great to at least hear something like “(quote) is where you lost me”.
@ChrisChiasson we talked about error tolerance at `expect_near_rel` behaves weirdly for values around 1e-8 and my proposals for change were merged in Improved behavior of expect_near_rel by martinmodrak · Pull Request #1657 · stan-dev/math · GitHub . Please note that the proposal I made here is completely orthogonal to how error tolerance is computed, the error tolerance discussion in this thread started from a tangential concern.