Some of the recognizers need to be improved, and to do that we need data, ie squigs (the lines you draw that get interpreted as strokes). The system is now rigged to give everyone vocabs with characters with strokes that need to be written more to tune our recognizers. Which means you can’t choose between difficulty or style of characters.
Some of the recognizers aren’t even built yet though, so no matter how many times you try to write a stroke, it won’t show up. When you get to a stroke that will not be recognized, just write it a few times, say three to five, and then just move on to the rest of the character. As we get enough samples we’ll build the recognizers for the strokes.
Also, we’d like a wide variety of samples, so it’s better to keep the phantom off when you’re writing the characters (and this also makes for better learning). We don’t want tracings of the stroke, we want as many different styles as possible so that the recognizers will learn to be flexible.