The notepad method relies on accurate range measurement and AOB measurement to get it right. During the speed measurement it only measures the amount the periscope turns while it is locked on the target. From those data it can calculate a knots value. While it is locked on the target, the AOB pointer is kept updated internaly as the lock moves. So that is not the cause
This method has to start with an accurate range. Unfortunately this is difficult due to mastheight errors in the recognition manual, but also due to rendering artifacts. (mast gets clipped from the image because it is thinner the pixels) Also, you need to put the lower line at the right waterline of the ship. It's easy to get the range wrong here. When you are on the surface you can ask your deck officer to provide a range and compare it. But that of course won't help when you are submerged.
AOB is also tricky to read because of perspective changes. It takes effort and practise to get it right estimating it by eye. But when it reaches 90 AOB then the differences are less problematic.
One reason of inaccuracy that can happen during the speed-measurement part of the notepad method, is slight changes in own course due to wave action. If you don't take enough time to allow the bearing to change significantly then the slight wobble in the lock might get it to give the wrong speed. The wobble of the periscope should be a lot less then the amount the periscope turns due to the moving target.
[EDIT] The moral of this story is: try plotting instead, to learn speed and course.
