The longer your reconstruction filter the closer the upsampled bitstream can be to matching the original data points (where provided). But it will always be "close, but no cigar", and I have strong reason to suspect that any attempt to postprocess the calculated output to match the original data points (where provided) would then be doomed to further distort the signal instead of making it better.
Indeed. And never forget that we listen to music with our ears, not our eyes. An upsampled bitstream that passes through all the original points may LOOK better on a graph, but would introduce distortion in terms of aberrant frequencies and/or frequencies beyond the Nyquist of the original signal (unless you had infinite computer power; see above)
On the other hand, if we look at the goal of Redbook upsampling as to try to guess at the ultrasonics that could not be recorded by the original CD, then the technology to look towards should be blind spectral band replication, not some magic way of connecting the dots:
https://www.google.com/search?q=blind+spectral+band+replication
You can see how the conversation got started in here. But I'm fine with a mod forking these discussions over to the Science forum.