May 1st, 2014
I am working towards making a new release (17.1) with some bug fixes and scoring improvements.
I have been running recent changes through my standard test suite, consisting of 32,000 automated fast time-control games, played against 8 different opponents. This runs in about 8 hours using three machines, including my 24-core server box.
Recently I noticed the ratings were improving nicely but then I realized that I had turned position learning on, to run a match geared towards tuning the opening book, and I had forgotten to turn this off. So once I turned it off and re-ran things, I am getting lower rating numbers but hopefuly they are more reliable and reproducible now. (This experience does show that position learning works, though).