Patrick, thank you for your kind note. I discovered, as you have known for a long time, that testing RTTY is not easy because of random figures/letters shifts. As you said, a single inappropriate shift can mess up a lot of characters! That makes the statistics difficult.
My test text file is at http://mysite.verizon.net/wz7i/Text%20file%20for%20testing%20communications%20software.html I used call signs and about 30% five number groups to try to deal with this issue. I tested with UOS off because of the number groups. It may be that I should have used a shorter file and then tested it with different audio files a number of times to get reasonable statistics but that seemed too much work... chuckle... The error bars on the graph might have been significant. Instead I tried to run a long enough text file to average out all the random shifts. It probably wasn't long enough to try to analyze the data too closely. I, too, tested with AFC off. I used the audio frequencies used for FSK so that is a difference. Our audio levels were about the same -- 40% sounds about right. As I said earlier, it is possible that I have incorporated some error in my methods. It is possible that I am "straining at gnats and swallowing camels" :-) Thank you for your patience with me. 73 de Wes, WZ7I