Debate: testing at fast time controls

[Kempelen]

So it takes a _ton_ of games to get down to +/-2 or lower. way more than I originally thought.

Those data are what make me feel annoyed because many people like me can't run so much games. I make little tests and follow my intuition which my only tools. I suppose many people is in the same situation like me......

[bob]

So it takes a _ton_ of games to get down to +/-2 or lower. way more than I originally thought.

Those data are what make me feel annoyed because many people like me can't run so much games. I make little tests and follow my intuition which my only tools. I suppose many people is in the same situation like me......

Almost all. I was doing that same thing up until last year myself. Unfortunately, I now see how many mistakes I was making as a result, also. I remember Ed talking years ago about having 8 computers, each pair playing a normal long game, so doing a total of 4 games at a time, 3-4-5 hours per game, and trying to draw conclusions after 100 games. And I thought "I wish I had the time and facilities to do that kind of testing here..." Unless the improvement was _enormous_ 100 games won't show anything... Everyone claims to understand the error bar for Elo measurements, and most probably do. But they don't understand exactly what it means, and how much variableness there is in the results while remaining under the error bar... It is too easy to focus on "the Elo number" by itself, which is misleading.