A few years ago I worked with Nam Seog Park on a mechanism for performing reactive inference, based on an original model by Shastri and Ajjenagadde. This has some useful properties but also some undesirable ones - the most obvious being that (although performing a symbolic style of inference) it looks "connectionist" in representation and is therefore difficult to engineer directly. However, we know how to translate from a suitably constrained symbolic specification into this lower level representation. This might make it an interesting final stage in the system of layered design which I am building now. I plan to introduce part of the new notation and explain how it might fit into the larger plan.