The #2 looks like it didn't autofeather until 3 seconds AFTER the uptrim on #1 - and there doesn't seem to be anything in the trace that precipitates the #2 autofeather?
Note that some parameters are only recorded every few seconds, sometimes as low as every 4 seconds, so if something appears to happen 3 seconds after another event, it may actually have happened almost 1 second earlier, but the change wasn't recorded until later.
A hint to the frequency is given by the slope of the linear segments combining the data points. I have long said that doing this kind of linear interpolation is wrong and misleading. These graphs should only show the actual data points and not try to imply data where there is none.