Solving pairs - proposal for possible improvement
Right now, you can make any number of mistakes in solving pairs of words and the answer will be considered correct when you finally manage to connect all the pairs at the end.
My suggestion is to change it to be like the following. When the user makes a mistake in connecting words, stop, select the first word the user has clicked on, and give a user single chance to correctly connect that word. If he succeeds continue as before, if he fails, then the answer is bad, and treat is like bad answers (repeat at the and of the lessons).
This way user can still be quick and carefree, but confusing words would be isolated better.
Hope you agree that this way it would be more interesting.
I'd not assume that making mistakes in pairing has no effect at all. Many people think that the software knows whenever one uses a hover hint and decreases the strength of the word in question commensurately, so it's very possible that a similar thing happens in pairing exercises.
What I'd most like to see changed about these exercises (on the web) is a means of making it possible to do them using only the keyboard (numbered boxes, perhaps); they can be a nightmare in timed practice (and your proposal would probably make this even worse).