The problem with nearest neighbour models is that they are not efficient in high dimensional spaces, because of the "curse of dimensionality". Basically it means that when a space has a lot of dimensions (features), your data points become very sparse and distant from one another. So when you input today weather into the system, it is far from every other previous points, because there are too many variables. Statistical models, or deep learning, will find invariants, regularities, etc... to fill that void and provide more accurate predictions.

You could also use this system in a reverse situation. You light a target, and subsequently try to hit it by sending shoot orders via the trigger. When you get all parameters right (winds, elevation,... ), you score a point, no round is ever fired. Maybe this already exist for training, I am not literate in that matter. You could also shoot blank rounds, should you need to take recoil into account.

If you use an open source client you could end up with the same problem. You can connect with Pidgin to gtalk, using SSL, and still have Google read your messages.