I went home in the late afternoon (around 5), showered and ate, and came back to the office in the early evening. The evening was full of tea (lemon-flavored black tea), good cheer, and only rarely a fit of mumbling confusion.
I'm comparing my simulated data to some measurements. I can run my simulations to convergence in order to feel some confidence that I'm getting the numbers appropriate for my model; but it is a model, and I know that I have the geometry slightly wrong and the material properties slightly wrong. And then there's the measured data -- which lacks any error bars. Being well aware of Murphy's laws, I expect that there are errors in the simulation from model simplifications, and also issues with the measurements that might have shown up if the number of measurements per device type was greater than one.
If science is a blind man exploring the world by vigorously poking with a stick, engineering sometimes seems like what happens when the blind man is handed a spear.