Forrest_

I spent a few hours trying to wrap my head around a Ginsberg paper on partition search for DDS and I never really grasped it, but it seems like that might be what is going on here. Paper The idea is that you can reduce the search space significantly by consolidating positions. For example, if the last two tricks are going to be won by AK of trumps, then the order everyone else pitches their final two cards in doesn't matter, nor does which final two cards they have come down to - so if you can identify that you've reached that set of positions (and you've already solved it once), you can stop your tree search. The part I never quite got is how you will know when it is safe to say that no decisions matter. I don't have any evidence to support this, but I haven't seen any claims that the DDS for GiB is broken when evaluating a single deal (e.g. through the BBO history UI). I wonder if the issue is caused by trying to reuse transposition/partition table entries across multiple deals when attempting to simulate the hand.