Here's the "masterplan"
: Analyze the raw data to hopefully find how to interpret it differently for the version 1.x. If that fails, then analyze the "glitches" and build a "high-pass" filter (something really simple) to "cut off" (skip) glitches. The last will remove extreme cases but sometimes we cannot make the difference between a glitch and a real stroke. So I am hoping to find the solid solution before trying to filter it out.
You might get lucky with lazynezumi which is doing that type of filtering already and a few members are using it.