It sounds like IIR is about reusing previous calculations? But you don't actually want every previous sample to affect the result; it's more a side-effect of how the calculation is done?