You are involved in a process. Occasionally the process gets "off-track", and you have to fix something to get it back "on-track". You want to minimize the time spent "off-track". Assume the time it takes to get back on-track is short compared to the time between when the process starts to drift and when you notice the problem. Therefore you want to notice the problem sooner. One possible approach is to borrow a concept from control systems engineering. Control systems engineers often build navigation systems that combine readings from a model with readings from a sensor to estimate position. The model (a real-time computer simulation) has errors, and the position sensors have errors. Combining the two gives a better prediction then either alone. A less formal application of this concept would involve using knowledge about a process's dynamics to improve when process observations were made, or how process observations were interpreted. It is inefficient to weed a garden every four hours. It is also inefficient to weed a garden only when you can see the weeds from your window. If you know about how long it takes weeds to grow, you can optimize your weeding schedule.