Ok, as I suspected, although still a bit frustrating. I think that most engine writers implement the first few pieces of the search/evaluation in a relatively logical order (although there may be some transpositions, to make a bad pun). But after that it just goes in whatever direction the writer wants to take it.bob wrote:I don't think there is any good "metric" for this. Some things are significant, like null move or reductions or search extensions, somethings are more modest such as different pieces of evaluation knowledge.
null-move will do a lot to speed up your search. But this really is an evolutionary rather than revolutionary process. Every little idea will make it better, but the ideas that work will come slower and slower as you get more things included.
Apart from the basics, I have modest quiescent search and check extensions (6 and 2 plies, respectively), MVV/LVA and PV move ordering, some very basic positional evals and 3-rep draw detection. Certainly nothing exciting for this board.
But then I get a position like this in a game and wonder if my engine performed poorly because I have done something wrong so far, or if it's because I haven't gone far enough. From your response, it sounds like the latter, which is more encouraging.
Thanks again,
jm