Test engine with positions

Answer

Question: I let the new Fritz versions "routinely" run various position tests with given positions.

There are always deviations in the recognition of key moves. Even if I know that a test is not the "panacea" - it does not seem to me like progress in the right direction.

Answer: With these methods, e.g. the simple position test, you cannot "measure" playing strength.

You have to repeat the test a statistically significant number of times to be able to make any statement at all. A single run gives useless values. 10x is better, 100x offers the user already quite reliable statements. And tactics tests do not give any real basis at all for evaluating playing strength. These tests only measure how fast exactly these positions can be solved. In order to be able to make any statements at all about tactical abilities, a statistically significant number of positions is also required. We think that a test with less than 1000 positions has little significance for the evaluation of an engine.

Tags
Created on
12.04.2023
Rating
Feedback

Back to List