An experiment - all oscillators off, filter emphasis set to self resonate, both keyboard control of filter switches set to off and small amount of contour - it's really easy to check the behaviour. It seems to be influenced by the decay time of the loudness contour as well, which is surprising, as I assumed the filter contour and loudness contour would be independent. With small amounts of loudness decay, the filter cutoff jumps around quite erratically when notes are played quickly. As the loudness decay is increased the cumulative change to the filter contour is smoother, more predictable.