That was a basic assumption I made. In an earlier post I proposed to the OP to apply some factor to the "promotion distance", which implies exactly what you say. I also think that cases where a king move has in fact a lower DTM than advancing the pawn (while both moves win) are not harmful since I would regard winning by advancing the pawn as "making sufficient progress". I'm not even sure whether many such positions exist, or any at all.hgm wrote:Only as long as King advance has a smaller bonus than Pawn advance.
No, we are only talking about which eval criteria to include for bitbase-won KPK positions to support progress. Moves advancing the pawn but resulting in a draw are out of scope. Bitbase-drawn positions get a draw score and nothing else. Our "positional" KPK evaluation is restricted to won positions only which we have to compare in order to quickly find a satisfying PV, i.e. one that results in winning by actually making progress (I don't think it's necessary to aim at the shortest path to mate, that would be a better task for an EGTB-based engine). Due to iterative deepening it does not matter much whether we do 1-ply or N-ply searches since each new iteration starts with the previous PV.hgm wrote:The main point was that there are also positions where advancing the Pawn draws. Then a 1-ply search would not know what to do if you did not pay attention to the King.
Only in positions where advancing the pawn draws but more than one king move wins it might become relevant to include king-related properties into the KPK evaluation. But I don't think it is necessary to be too sophisticated in this case since the search will quickly find those king moves which later on lead to a winning pawn advancement as early as possible, and I think it would even be able to do so without any king-specific term at all.
Sven