Uri Blass wrote:If the depth is big enough the engines do no mistakes even without tablebases so tablebases give no advantage.
When the depth is small the engines may often blunder even when the root position is tablebase position and they may often fail to win KRP vs KR or KQP vs KQ because of lack of tablebases.
You apply human chess thinking to engines which is just wrong.
Word "often" that you mention doesn't mean a thing. Also looking into specific endgames simply means nothing (these situations are far too rare to count for anything).
TB advantage is all about statistics. You can see TB access advantage in 2 ways, reducing necessary search time (since from that node where TB is accessed there is no more search deepening, the score is final, it's like a hash entry with infinite depth) and having less faulty evaluation (in nodes where TBs are accessed). Of course there is a penalty in loading a particular TB in memory and decoding a position, but once you start doing it at high enough depth (meaning there will be more TB hits) the penalty is minimized.
The advantage of TBs translates directly to how many nodes out of total nodes have been evaluated in TB's. The more there are, the bigger TB advantage is. And this is directly proportional to the depth reached.
This is pretty straight forward (I do realize that for ppl who are not chess programmers and still think in terms of particular endgame positions this is quite hard to understand) but would expect you to think more as a programmer and less as a chess player...
