bayes for days

docs(features): balance the los_in/los_out pair and refresh the stale notes master

The trace is only asymmetric where heights differ, so the pair cancels on flat ground. Measured over a real 8v8: `los_in` and `los_out` were identical on all 2985 candidates, so unequal weights would have made the pair a net "being visible is good" term that neither sentence says.


+69 -12
5 changed files