not giving any guarantees, but have you tried the logarithmic mode flag? that might trump the linear auto-in-line-smoothing-simplify-optimization-thing :)
be warned that the "auto-simplify" feature is most welcome and convenient when dealing to recording automation, especially if the target subject is mapped to some live tweak-able MIDI knob/slider controller ;)
basically it tends to drop redundant automation data point nodes, ones that add nothing but cruft to the curve slope. it has its short-comings namely that it only does it by just linear regression.
the case on hand is about you visual editing the values via mouse click-n-drag, whatever--point is linear moves on the screen are directly proportional to scalar value changes, and that applies 1:1 to linear and spline modes while (sample&)hold mode is special, obviously.
now say, if your automation curve is set into so called "logarithmic" mode, as is found convenient for dB valued data series like gain etc. you'll be detaching the linearity predicate between actual data signal and the visual graphic where you move things around.
not giving any guarantees, but have you tried the logarithmic mode flag? that might trump the linear auto-in-line-smoothing-simplify-optimization-thing :)
be warned that the "auto-simplify" feature is most welcome and convenient when dealing to recording automation, especially if the target subject is mapped to some live tweak-able MIDI knob/slider controller ;)
basically it tends to drop redundant automation data point nodes, ones that add nothing but cruft to the curve slope. it has its short-comings namely that it only does it by just linear regression.
the case on hand is about you visual editing the values via mouse click-n-drag, whatever--point is linear moves on the screen are directly proportional to scalar value changes, and that applies 1:1 to linear and spline modes while (sample&)hold mode is special, obviously.
now say, if your automation curve is set into so called "logarithmic" mode, as is found convenient for dB valued data series like gain etc. you'll be detaching the linearity predicate between actual data signal and the visual graphic where you move things around.
have i made it a too complicated excuse? :)
hth.
cheers
anyway