Help
West chose to lead the 8♦ at MP.
At first sight, nothing seems odd about this. Everybody knows that GIB leads passively against NT. And in cases where it's close, it still often gets it wrong because it doesn't run enough simulations, or its method at deciding what hands N/S could hold are bad, etc etc. What's new?
But running my own simulations, Q♠ was coming out on top every single time, while when I ran it with GIB, a passive minor was on top every time. Why the difference?
It's due to a very simple bug. (Well, two actually, but one is old news.)
GIB is programmed to generally follow lead conventions, like 4th best. This is done after the double dummy scores have been calculated, by adding a small bonus to 'book' leads for each suit combination. At MP, I believe this bonus is 0.2 of a matchpoint.
The old-news bug is that the 'book' lead from QJ92 is the 2, when every human would lead the Q. This has come up in past threads - not by it actually making that lead, but later in the play when GIB in another seat is simulating which hands opening leader might have held to make that lead.
The new bug - these bonuses are being used for making comparisons across suits, when they should only be used within the same suit.
That is, suppose GIB comes up with the following scores for each card (these aren't the actual numbers, but I adjusted them slightly to make the point clearer):
- ♠Q scores 0.14
- ♠2 scores -0.1
- ♦8 scores 0.07
- ♦6 scores 0.07
If you're going to lead a spade, it may give the ♠2 a bonus, but should still pick the queen because it's significantly better.
If you're going to lead a diamond, it will give the 8 a bonus, and lead that.
But here it sees that ♦8 with a bonus to be better than ♠Q without a bonus, thus leads a diamond.
Putting it simply, it's preferring a diamond not because it thinks it is a better lead, but because it would rather lead the wrong suit than have to break its rule about 4th best.
Myself included..