You are much better at using all these parameters than me, despite the fact that I designed them! If You want, I can leave designing personalities named after real players entirely to You. If You would like to prepare a package of such personalities by mid-December, then it will become part and parcel of official Rodent III release, with credits and all. You can safely use 0.212 version to that goal - right now I'm trying to tune eval parameters automatically, but personalities will keep using the current set of values.@Boban, I know that endgames are tricky and that Rodent becomes a bit weaker towards the end of the game. But coding them is even more tricky. First of all, I'm not a very competent endgame player. I know a couple of "can't win that" patterns, and coded them into Rodent, but my engine has endgame problems way earlier than these patterns can influence score.
Perhaps the general problem is that Rodent's evaluation functon is skewed towards the middlegame. It knows quite a bit about positioning minor pieces (defended, in a hole of enemy pawn structure, able to reach such a hole, attackable by enemy pawn, hidden behind own pawn, restricted to own half of the board, special code for fianchetto and for a bishop returning to f1 after white king made a short castle...) In the endgame this knowledge becomes meaningless, and apparently all the details of passed pawn evaluation are not on the same level.
Another issue is adding endgame knowledge. Coding one pattern is essentially easy. But after the pattern is added, search becomes really creative about finding similar positions that do not match this pattern. So it is not enough to block one way of going wrong - one has to find basically all of them.
Also, basic human intuitions surprisingly often fail to work when translated into code. Recently I tested a simple patch that said "pawn endings should be scored to 110% of their value", so that Rodent would prefer to defend rook endgame (or a minor piece endgame) while a pawn down. And it weakened the engine.
In short, I do not know what to do about the endgames. It might be a search issue. Rodent might need a different set of pruning/reduction margins for the endgame. Or some of the weaknesses might be fixed by working on that little specialized move generator that is used on the first ply of the quiescence search. I simply don't know.
