I checked 2 million sentences from my text collection. These sentences were selected, because there was no spelling error in them. Nevertheless, out of 2 million sentences, LT reports almost 1 million times.
That is quite a lot.
With the collected errors in a database, it is easy to check the validity of detected errors, so there is some tuning to be done.
But I also know there is still quite a lot that goes undetected. I am thinking of ways to improve that too.