I add improvements any time I feel the benefit the program. I keep a file for people to read at the group called ongoing programming. In the file I've tried to explain the difference between an improvement versus a disaster.
I have run into people who feel there is no change except to fix a disaster. At some point they will have to confront the real-time test, wherein I have been making program improvements all along.
I'm avoiding discussing any results before the real-time test. I feel it can and will make the same statement but will be based on publicly recorded data, rather than a backtest.
I forgot the one thing I am encouraged by: the real time signal generation and result so far is tracking dead on to the backtest tradelists I have in the files section. This is a promising sign.
Also I forgot to mention that signals for all of 2007 are in the tradelist section.
Market Data 
Markets 
AI
