The article introduces a new way to estimate important variables in systems that are hard to observe fully. They use a special kind of neural network called a deep filter to figure out these variables without needing to know everything about the system. This method gives a number that shows how well we can observe these variables in different places.