I've been doing that sort of thing with my LS program, too. A large blunder will show up with some candidate measurements to recheck, but the distribution of errors does blur things somewhat.
When you have poor values or multiple small blunders, so the original data didn't mathematically close in multiple places, I guess the issues of what to hold and what to let move would be little better or worse than using cogo.
---------
The lack of original closure has plagued me on the two subdivisions I've tried to compute. For one of them I entered all the distances and bearings (where given) for 350 plat points and there are discrepancies of up to a foot all over the place. Generally lines tend to be longer than their surroundings allow. Guessing where something might be blundered and freeing that measurement never seems to make the rest fit either.
I suspect they used slope measurements to lay out the subdivision with the design as a "guide" and put their closing measurements on the final plat. Have you ever seen that done?