okay so your request for the pose based transform id suggest the request stand something like this (being exact showing mathematical principles is faar more likely to be cosidered by whoever it is gets to do the assignment, otherwise its no guarantee you get what you want)
instead of having the blendshape allow only for additive blends. like this:
blend1*blendvector1 + blend2*blendvector2....
It would be good if it could also yield the weighted avreage, max and maybe a smoothsetpped max if possible.
weighted average would be:
(blend1*blendvector1 + blend2*blendvector2....)/blend1+blend2...
tested this and divison by min(blend,1)+min(blend2,1) is also quite usefull maybe smoothstep it somehow.
max would be
max(blend1*blendvector1,blend2*blendvector...)
tested and yes its the max filtering that is causing the popping hende the time based smoothing. just as i expected.
however i have no idea how the smoothmax should work. Probably by storing a vector of change, that does not permit faster change that that somehow. or comparing it to the next higests current and last value perhaps (this is bad since taking last frames info is either very bad in terms of maya usability or ist going to make the plug a nonscrubbable sim wich i doubt no one wants)? Perhaps a loop node would solve this?
il test this later, now im off to do some advanced calculus.
As for the gaussian blurring of movenment in between... am still not covinced we need this. But hey what do i know. relatively easy just create some extra internal blends that weight based on a gaussian(sinc?) distribution multiplied by the entire blends driver. Is relatively easy to script so i would keep it in scriptmode as it allows for more flexibility, against usability for beginners offcourse.
While were at it having the possibility for storing the vector changes in UVN space as well as object one.
THe blend is realtively easy to make trough the api, so im not so sure its neccesery for alias to do this. But its allways nice, unless somebeody is willing to code the plugin and release the code.
Additionaly youd want a intteligent datadriver that easy to manipulate for pose driven stuff. Any propositions on how it should decide what is the driver for what apart of the body? if its not super automated with complex heurestcs again i would look at mel to do this but again its your taste.
okey skip all thats said! true enough maxing (but maxing what 
) is the way but how its implemented is no longer a complete mystery, and true enough i dont need to blur the info in 2 of the solutions ive come up with (mind you ives pent 2 hours on the problem), im now building a aprticle operation that does this (wich gives me free per point blend option if i wish to opt for it). Will be postion more on ths as soon as i get the exact formulas sorted. The guess ahead and use the largest for offsetversion seems to be quite good, solution. Its computationaly doable and needs only to know odf current state and gives treshold.
