Currently there are only 2 options for selection of mid/side coding for stereo audio, no mid/side and adaptive mid/side using fixed 2nd order prediction.
The plan is to first add a framework and option for exhaustive mid/side selection. Then, other options will be added such as constant mid/side and more complex estimation. Some ideas for better estimation that I might try would be using fixed 4th order instead of 2nd order prediction and trying a higher Rice order like maybe 4 or even the max order instead of 0.
I plan to have 2 options, one for enabling/disabling mid/side and another for mid/side method. This goes along with a general design change to separate bitstream constraints from algorithmic complexity.