IWB wrote:6. Indeed H5 has 3 losses on time in 480 games. That is below 1% ... overall. All 3 cases would be draws. It lost 1.5 points out of 480 possible ... not even worth mentioning it. But I will keep an eye on it!
That's unexpected, if the number increases further I will have to make an update to add a bit more security in the time management.
As for the "200 Elo" improvement claim, that's a measure in the best case scenario (self-play with 1 thread at super-fast TC).
It's rather difficult to produce a meaningful estimate of the Elo difference when the test conditions can vary immensely (TC, opponents, number of threads). In the TCEC chat I've always been reluctant to quote a value, it just doesn't mean a lot. Then again, the first thing people always ask is "how many Elo stronger"
If the measured results by the rating lists are too different from the "marketing" 200 Elo value, I'll definitely update the web page; maybe 150 Elo is more accurate in a real situation against many opponents at normal time controls.
Thank you all for your support,