I disagree. If you have an engine with a very light evaluation function and a very powerful search (i.e., it takes its strength from tactics, not positional knowledge), it can -very- time control sensitive, but not badly written.Gabor Szots wrote: ↑Mon Sep 28, 2020 7:54 pm Christophe Théron, author of Chess Tiger said once: if an engine is sensitive to the time control, then it is badly written.
If you play a game between two engines, one with a super-light evaluation and powerful search, and the other with an evaluation that encodes half an encyclopedia, then they might be equal at 10 min/game... but if you go to 20 min/game, it is very well possible that the faster engine will gain MUCH more depth using this extra time as compared to the slower engine.
The slower engine won't get a lot stronger with longer time controls, but the faster engine will (i.e., the longer time control benefits the faster engine more). If I remember correctly, around the time of Fruit, "fast and light (on eval)" was the norm for chess engines. It's only recently that "slow and heavy (on eval)" is becoming a thing again, with the neural net engines.
It is the reason why adding more knowledge in the evaluation is not always helpful: more knowledge slows the engine down, and the slowdown can cost you more elo than the extra knowledge gains.
That is why I'm convinced that:
- Engine strength is mainly gained by speed and a powerful search.
- Engine personality and style is gained by the evaluation terms. (Obviously more and better evaluation terms bring more knowledge, and if this knowledge gains more Elo than the slowdown costs, the engine will become stronger; but I believe the main part of the strength is in the search. I think Ronald's PeSTO has already proven this clearly.)
