This system doesn't parse any differently from how you'd expect for {ji'i xa pi xa}, as in the linked example you gave, but simply allows for constructs like {pa ji'i xa} for "16, with 6 being uncertain" -- clearly intended for simplifying significant figures and error analysis. This is the behavior I grabbed
straight from the cll. The "rounded" semantics is implied, but not the "rounded up/down" semantics. (Which I'm perfectly fine with.) What I'm not fine with is that {ji'i PAI} makes logical sense, and sometimes {ji'i DUhE}, but these are currently disallowed since they would put {ji'i} in a separate pagselma'o. I wrote a possible work-around in the google doc definition, but I don't particularly like it. Given, it's basically the same sort of thing I did with {pi DUhE}...
I've been trying to match the current system as much as possible, primarily disliking {ce'i}. I didn't know there were previous thread expressing dislike with {ji'i} -- personally, I'd like to see both interpretations able to coexist.