Reading the .mtf is most of what scoring one costs, and an application asks the same design many questions. A held design keeps its base per condition and the crew only scales it, which takes re-crewing from 432us a unit to 1us - the property the native benchmark promised and the boundary was throwing away. The smoke test checks both paths agree to the point, because the multiplier is now applied in JavaScript. Change-Id: I7e4fa6e18e2ab26e01bf035f7dcfbfc8e168426c