Mastering the Robots
(Page of 9)

We've been playing in the ACBL individual robot tournaments recently, and slowly learning our way.  Now, we are not  programmers, and don't have any real GIB expertise, but we are beginning to get a feel for the strange robot foibles, particularly when they are defending.  Like any other declarer-play exercise, you place cards, you figure out why they are doing X or Y, and you find a way through the tangle to make your contract.

This simple suit combination taught us quite a lot about the program.

North
AQJ10983
South
72

We cheated a point or two, and opened 1NT.  Robot North couldn't take a joke, transferred, and checked on key cards.  Missing two, we stopped in five spades, which looked safe enough.  Unfortunately, Robot West led the club ace, and gave Robot East a club ruff.  So, we had to pick up trumps to survive.

What is the best way to play this trump suit, once East has trumped a club?

That's a pretty easy problem.  If the trumps were originally 2-2, then either West started with Kx, and we should finesse, or East started with Kx, and we should play for the drop.  That's a 50-50 guess.  But, of course, if West started with three trumps, we must finesse, so the finesse is clearly the percentage play.

Or so we decided.  But our choice was moot, as the king popped up on the first lead, and we claimed five.

It wasn't until much later, when we looked at the hands, that we found out trumps had been two-two all along.  Wow!  West had hopped king from Kx.  Why???

Here is our hypothesis, and it is two-fold:

(1) The bots analyze play positions double-dummy.  If a finesse is onside, they know that you know that, so that card becomes useless.

(2) If bots drop the highest cards from equals, stupid humans, who aren't counting, may lose track, and the bots gain a trick.

Humans might well be aware that the missing queen of clubs is still out, but they might not notice if the only club out is the high five-spot.

Is the hypothesis correct?  Over and over, we have seen the robots follow this silly defensive strategy.  And it certainly explains the craziness on this hand:  The spade king was doomed, so why not play it?  Maybe the stupid human declarer will forget that there is another trump out.

Back to our suit combination problem.  Suppose the computer had followed small.  Should we still finesse?

Not clear, but I suspect no.  The computer would not fly king from Kxx, since that might be worth a trick by force (when declarer has only one trump, and can only finesse once).  But Kx offside is more likely than a singleton offside, and so ...

Here are some fairly easy play problems from recent sessions.

(1)

North
Q84
A7532
AKQ6
Q
South
A972
J6
8752
KJ3
W
N
E
S
P
1
P
1
P
2
P
2NT
P
3
P
3NT
P
P
P

West leads the club six, and the queen wins the trick, as East plays the nine.  This is typically a standard count card, so you can place West with A10xxx in clubs, and East with four.  You play four rounds of diamonds.  East discards the club four on the third round, and the spade five on the last diamond.  West, who started with J109 in diamonds, discards the heart king!

West
North
Q84
A7532
AKQ6
Q
East
South
A972
J6
8752
KJ3
W
N
E
S
P
1
P
1
P
2
P
2NT
P
3
P
3NT
P
P
P
D
3NT South
NS: 0 EW: 0
6
Q
9
3
1
1
0
A
3
2
J
1
2
0
K
4
5
10
1
3
0
Q
4
7
9
1
4
0
6
5
8
K
3
5
0
5

These cards remain:

North
Q84
A7532
South
A972
J6
KJ

OK, who has the spade king?  Why?

East has the spade.  West has shown up with five clubs to the ace, the heart king and queen, and the diamond jack, yet passed originally.

Even if West were not a passed hand, the spade king is marked.  Remember, analyzing double dummy, West has no reason to duck smoothly in spades.  West knows that you know where the king is!  The bots almost never duck in these situations.  Their algorithm shows that you will win the queen, and they may never get that trick.  So, they grab it.  On the rare occasions when they duck, they have to run their simulations, and that takes time!  So, they never duck in tempo.

You need a second spade trick to make the contract, so you will need to find East with a doubleton spade king, or West with a doubleton spade jack or ten.  That latter is more likely, and, anyway, East has discarded a spade, which is very unlikely from Kxx.

In practice, this intrafinesse works just fine, and you wrap up 430.

(2)

North
AQ10753
K72
A7
103
South
A108543
8654
AJ9
W
N
E
S
P
P
1
P
2
P
4
P
P
P

West leads the trump six.  Plan the play.

If trumps are splitting, there are ten easy tricks, so it is obvious to worry about trumps 3-1.  Could the lead be a singleton?

Yes, robots love to lead singleton trumps.  Robots often go passive on opening lead, and a trump lead is the best kind of passive - gives away nothing, and may stop ruffs.  Of course, leading a singleton trump may pick up partner's holding in the suit, but not in the double-dummy world of robots.  You knew the trump position anyway, so would always guess the suit.  I've seen robots lead the singleton king of trumps.  Hey, why not?  It could never win a trick anyway, right?

Just as the bots love to lead singleton trumps, they hate leading from a trump honor, which, in their simulations, could easily blow a trick in the suit.  So the lead is most likely from the singleton six, or 96 doubleton.

We are still in great shape if there is a trump loser, so long as a club honor is onside.  We can easily win two clubs, five trumps, two aces, and a diamond ruff on the table.  Ten tricks.

That makes the start obvious - win the trump king, and start on clubs.

So, you win the trump king (nine from East), and play the club ten, to the eight, nine, and queen.  Back comes the diamond queen.

West
North
AQ10753
K72
A7
103
East
South
A108543
8654
AJ9
W
N
E
S
P
P
1
P
2
P
4
P
P
P
D
4 South
NS: 0 EW: 0
6
K
9
3
1
1
0
10
8
9
Q
0
1
1
Q
3

Over to you:

We more or less decided that trumps were 1-3 at trick one, and that seems confirmed.  So we have a trump loser.  One of our losers will go on the spade ace.  We might, in theory, trump the other two losers, but that means giving up a diamond, and East will get in, to continue trumps.

Seems like we are forced into a second club finesse, and may as well win the diamond ace before finessing.  If successful, we'll make five.

That finesse, by Restricted Choice, is a two-to-one favorite, right?

I don't know - I haven't collected enough data to judge this yet.  Early on, the robots tend to signal accurately, for their robot partner.  From their perspective, winning with the king, and trying to fool you, is useless - you already know about the club honors!  But robot partner might be fooled.  So, I suspect, at trick two, robot will almost always win the queen from king-queen.

However, there is another factor, one that applies to robots or humans.  Any bridge player, when that club ten was led from the table, would cover with an honor, or at least give it some thought.  Forget Restricted Choice.  The play so far strongly suggests that the club king is offside.  But, if that is so, we can set up our cross-ruff without letting East in to play a second round of trumps.  The right play, and the winning play at the table, was to play a club to the ace, then lead the jack of clubs, discarding the diamond loser.

(3) You pick up this pretty hand, both vulnerable:

North
A854
AKQ3
J9
A106

Playing standard methods, this is an automatic one club opening.  However, starting with one club is the worst possible choice playing with a robot.  They will pass with any weak hand, regardless of shape.  Starting with one club will see you declaring in one club, facing a hand like KJxxxx xxx xxx x.  Going down two or three tricks in one club, cold for five spades - not so attractive.

Never open one of a minor, planning on showing 18-19 balanced later, unless that suit rates to be a reasonable strain facing a weak hand.

Playing bot bridge, you have two choices - downgrade and open 1NT, or upgrade and open 2NT.  Either is reasonable, and both are vastly better than opening one club.

We chose to downgrade, and the auction went smoothly after that, to four spades.

West
North
A854
AKQ3
J9
A106
East
South
J762
2
AK2
98752
W
N
E
S
1NT
P
2
P
2
P
2
P
4
P
P
P
D
4 South
NS: 0 EW: 0
4
6
J
5
2
0
1
K
7
3
A
1
1
1
2

West leads the club four, and you see that you are in a normal, but poor game contract.  Looks like two trump losers, and two club losers.  Oh well.  You duck the lead, and win the club continuation.  What next?

Looks like we need a very luck trump lie - KQ doubleton, or a huge defensive error.   Or do we?  Maybe the hand will look like this:

West
109
J8654
Q874
43
North
A854
AKQ3
J9
A106
East
KQ3
1097
10653
KQJ
South
J762
2
AK2
98752
W
N
E
S
1NT
P
2
P
2
P
2
P
4
P
P
P
D
4 South
NS: 0 EW: 0
4
6
J
5
2
0
1
K
7
3
A
1
1
1
2

If the cards lay this way, the contract is cold.  Simply play some winners first.  When the fourth heart is led from the table, the ending will be:

West
109
J8
Q
North
A85
3
10
East
KQ3
10
Q
South
J762
9
D

There is no defense.  East will likely discard the diamond, as you ruff.  You exit your club, and will hold your trump losers to one.  Nice!

Of  course, the spade honors rate to be split.  Can we make the contract against the same shape, with split spade honors?

Sure, the ending would be:

West
K10
J8
Q
North
A85
3
10
East
Q93
10
Q
South
J762
9
D

Trump the heart, and play spade ace, spade (or spade ace first).  West must win, and give you a ruff-sluff, or unblock in trumps.

In practice, hearts were 4-4, diamonds 5-3, leading to the same ending:

West
K10
J854
Q8743
43
North
A854
AKQ3
J9
A106
East
Q93
10976
1065
KQJ
South
J762
2
AK2
98752
W
N
E
S
1NT
P
2
P
2
P
2
P
4
P
P
P
D
4 South
NS: 0 EW: 0
4
6
J
5
2
0
1
K
7
3
A
1
1
1
9
5
A
3
3
2
1
K
4
J
6
3
3
1
2
7
4
10
1
4
1
A
6
2
4
1
5
1
K
7
2
5
1
6
1
Q
9
8
J
1
7
1
3
10
2
8
3
8
1
6
10
A
3
1
9
1
5
11

We had some company at +620.  Two declarers won the first club, and played trump ace, trump.  Both West's won the king and shifted to diamonds!  One tried low, and declarer guessed well to put up the jack.  The other West robot got out with the diamond queen, so there was no guess.