Observation

[Rebel]

When I (also) moved into the circus of playing thousands of games some 1½ year ago to proof or disproof a change I had this habit to start another match right after one was finished, WITHOUT a reboot. This behaviour (often) could go on for longer than 4-5-6 days before I did a reboot.

During that time I (sometimes!) noted extreme fluctuations in match scores. A change could give a match result of 52% and even sometimes 53% and when repeated after a REBOOT there was no improvement at all.

After I noticed this phenomenon for the 4th time about 2-3 months ago I decided to reboot the PC after every match. And so far every match result is repeatable within the reasonable fluctuations one might expect.

I am using Windows 7, cutechess-cli for bullet games (40/15) and Arena 3 for longer time controls (40/60). I am assuming that the PC's memory gets corrupted while (Windows 7) time passes and starts to influence the matches.

It's impossible to fingerpoint to a guilty one or ones, just want to share.

[velmarin]

Tests of two engines, start first or second gives me distortion.
Clean the cache of the computer, with a free program JetClean.

Still, today you have a breakthrough, and the next test is back.

Now I'm doing tests Tournament, put two or three identical engines, in different order, against rivals, if it is against another version of the same engine, also two alike.
So I see something else.
The results seem more credible.

[elcabesa]

I'm having the same problem here, still investigating.
I used 6 seconds games and sometimes I get very unpredictable result. Mabe it's pure statistic aiation, but I can't understand it