[mike777]

xcurt, on Apr 30 2008, 09:15 PM, said:

mike777, on Apr 30 2008, 08:58 PM, said:

hmm are you saying over a strong nt 6s is 80% just looking at our responder hand, deal two, and before any other bids other than 1nt?

If so why did not one expert say this, not one?

Sorry I mistyped. Too tired. I meant "it's right to be in 6S 80% of the time facing a random strong NT."

Slam is making about 65% of the time if the opponents lead randomly from among the three offsuits.

The conclusion still holds though, partner won't move enough after 2H... 4S for 2H... 4S to be the right responder action.

Bashing is about a wash if the opponents lead randomly against the 2H...4S auction. It's going to be a gainer if the opponents can exploit the information gained from the auction.

Bidding slowly (How? Thats my point -- the action over 1NT is the real problem!) could be a big gainer over 2H...4S if we can avoid the 20% of the hands where slam has no play on any lead or is off two cashing tricks.

Ok, you say across from strong nt opening 6s over a strong 1nt is a big fav, more than 51% My main response is not one person in the article said this. This seems like a big thing to not mention for us nonexperts.

If true, and I do not mean to doubt you, it would really help to tell us, teach us how to know that. Trust me alot of readers will not know this.

[xcurt]

mike777, on Apr 30 2008, 10:59 PM, said:

If true, and I do not mean to doubt you, it would really help to tell us, teach us how to know that. Trust me alot of readers will not know this.

Sure, I'm happy to talk about this. I apologize for the long post but I want to express my thought process.

To attack problems like these, first get a copy of a hand dealing program. I use Hans van Staveren's venerable dealer program (see http://www.xs4all.nl...dy_dealer.html). If you are not a *nix guru you probably want to try one of the more recent Windows programs. Dealer can write some ugly output and you will probably want to hack up some perl code or similar to generate input files and work with the output.

All of these programs allow you to randomly deal some or all of the cards, select hands based on conditions, and write the hands or compute statistics over the hands.

Now you get into the philosophy of simulation. Personally I think that double-dummy solver analysis while useful is not the most direct route to the truth of a hand. I prefer to do two things.

First, calculate some statistics about partner's critical holdings. Here, we can simply predeal ourselves the 13 cards we see. We don't need to impose any conditions on the opposing hands since they are probably so weak they will never bid without extreme shape. We impose a strong NT condition on partner's hand. There will be some edge cases where you look at the hand and say "I would not open that 1NT!" If you see such a hand disregard it. It will probably be infrequent enough not to distort the statistics too much. In this case, we have a bidding problem in part because our high honors are not where partner will most likely expect them -- in our long suit. We also have a bidding problem because we have a suit with two fast losers. So we need to find out at least two things to bid slam with greater accuracy than just counting points or just bashing. We need to know that they can't cash the CAK, and we need to know that they don't have the CA and some trump holding that is likely to produce a trick. If we ask the dealer program how often partner has only one key card for spades, we find out that it's about 15% of the time. So while this is a problem we would like to resolve in the auction, it might not be the most important problem to resolve.

Second, deal partner a number of hands consistent with the auction and look at each one and think about where we would like to end up and whether or not given actions from us will produce the desired result. We need express the results in terms of IMPs or matchpoints against other tables that take other actions on the same hypothetical deal. I have found that this is very effective at helping me understand what is going on on a deal. In the case of this deal, I looked at 40 hands, single dummy. On each hand I looked at how would I play the hand. This was pretty easy since with a big 1-suiter there usually aren't that many lines of play. If you are looking at something like

Ax, KJxx, Qxxx, AJx

you would say on a non-club lead I would finesse the diamond first to set up a club pitch if it lost, then try the spade. You can figure out that you play this spade combination for 0 losers about 20% of the time, 1 loser 50% of the time, and 2 losers 10% of the time since you have too many trumps to pick up KTxx on your right with a coup. If the diamond finesse fails you make 20% of the time (edge case of 6-1 diamonds), if it wins you make 20% of the time plus 25% more of the time (spades for 1 loser and hearts to pitch the club or some kind of endpotision). There are a few additional chances in the endgame here like DKxx on and you might not need the HQ on. Since the edge cases will cancel out a little you don't need to be totally exact. So we write 37.5% for this one. We also have to estimate the chances partner would cue 5C with such a hand. If we always think partner is moving over 4S we have probably done something wrong since most of the time the auction goes tranfer--accept--game in the major it plays there.

Having done this for 40 hands or however many you think has given you a feel for how things are running, you now add up the probabilities. Some of these probabilities will be conditional (eg make slam given that partner has a cueing hand over 4S) and you may need to estimate the conditional probabilities. It helps to know something about Bayes Law here.

On the problem hand it becomes pretty clear quickly that when slam is bad it still has some chances (unless off the CAK or two aces -- which we know is only 15% of the time from a 1 million hand simulation so we know that number pretty accurately since the computer will count it for us). If slam is good it is usually cold or on one of two finesses or on the wrong lead or a finesse. It's good on most of the hands. So the real crux of the problem becomes how often partner is moving over 4S. I found it too hard to accurately evaluate each hand (judgement becomes colored by knowing the responding hand) so I considered what would happen if partner moves exactly half of the time. That's probably an over-estimate and you could answer that by going through a database of hands from BBO or OKB and asking how often partner moved over this auction. We now work through the algebra computing our IMP expectation against bashing knowing that

50% of the time one table is in slam and the other is not
50% of the time
85% of the time we have enough key cards and both tables are in slam (I ignore the increased
expectancy of having all of the key cards since partner moved)
15% of the time one table is in slam and the other is not

I won't reproduce it all here but you get something that suggests that even if
(a) the slow auction never bids a hopeless slam -- and some of the slams are hopeless because partner has all the missing quacks and we have two slow losers) and
( the opponents lead no better than against the bashing auction
that you are going to break roughly even by bashing if your only other choice is to bid as given in It's Your Call. In practice both (a) and ( are not likely to be as optimally distributed for the slow auction, so bashing should have a positive IMP expectancy against the slow auction.

Finally, there are other bidding plans out there. I would have bid 6S at the point given in It's Your Call, but at the table I would not be in that particular pickle. I really wish this had been given as a MSC multiple choice style problem after 1NT-?.

Curt

[mike777]

1) I have to admit I am sure this is all brilliant
2) I have no idea how this helps the mass of readers of this column at the bridge table under pressure to bid 6s direct over 1nt.
3) I am looking at deal one now. I would bid 4h since I think my hand is garbage but I see Larry Cohen, Lawrence, Jill, and Stansbys/ Freeman bid 4d. I think the write is up is bad but then I think it would take 20 pages or more to do a decent write up. In other words....great problem.
4) If cuebid is 100% with anything less than a dead dead minimum I understand 4d. But this is a big if and what is a dead minimum here.

[awm]

So your simulations indicate that with ♠QJxxxxx ♥AT ♦AJ ♣xx, you are making slam more than half the time on hands where opener would not move over the quantitative 4♠ invite?

I find this somewhat hard to believe. Surely with only 28 hcp between you or so, you could easily be off two aces or an ace and the trump king, or the club ace-king, etc. None of these hands land in slam when you bid quantitatively (you might get to 5♠ on a few) whereas all of them are in slam when you bid 6♠ directly.

My count is, over 30 hands:

6♠ cold: 4
6♠ better than a finesse, not cold: 4
6♠ on a finesse: 5
6♠ worse than a finesse but some play: 11
6♠ no play: 6

This suggests that blasting 6♠ may well lead to going down more than it leads to making. If we assume that the quantitative approach bids the "better half" of the slams you can see that it's a huge winner over blasting slam. More realistically, even if the quantitative approach just keeps you out of the "no play" slams and randomly gets you to slam on half the other hands, it seems substantially better than blasting.

[cherdano]

xcurt, on Apr 30 2008, 07:44 PM, said:

Furthermore, when partner moves over 4S, assuming he bids Blacke when he has all off-suits controlled and makes his cheapest cuebid otherwise, the relative frequency of his calls is

4N -- 20%
5C -- 70%
5D -- 8%
5H -- 2%

What kind of assumption is this? Usually partner will pass or bid 6S, only when he has a specific concern (lack of keycards, one suit uncontrolled etc.) but otherwise good cards will he bid anything else.

Where did you get your percentages of 6♠ making from, btw?

[xcurt]

awm, on May 1 2008, 01:26 AM, said:

So your simulations indicate that with ♠QJxxxxx ♥AT ♦AJ ♣xx, you are making slam more than half the time on hands where opener would not move over the quantitative 4♠ invite?

My simulation suggests that if we bash slam we will make it 65% of the time assuming the opponents lead randomly from among the offsuits. That introduces a free parameter -- how good are the opponents on lead. This parameter has a dependence on our auction. The more information we transmit, the better they do. The 65% figure is across all hands.

awm, on May 1 2008, 01:26 AM, said:

I find this somewhat hard to believe. Surely with only 28 hcp between you or so, you could easily be off two aces or an ace and the trump king, or the club ace-king, etc.

This is measurable. I measured it. Partner has at least three of the CA, CK, SA, SK 85% of the time. That's from a simulation with a million hands generated including roughly 40,000 where partner has the strong NT.

awm, on May 1 2008, 01:26 AM, said:

My count is, over 30 hands:

6♠ cold: 4
6♠ better than a finesse, not cold: 4
6♠ on a finesse: 5
6♠ worse than a finesse but some play: 11
6♠ no play: 6

This suggests that blasting 6♠ may well lead to going down more than it leads to making. If we assume that the quantitative approach bids the "better half" of the slams you can see that it's a huge winner over blasting slam. More realistically, even if the quantitative approach just keeps you out of the "no play" slams and randomly gets you to slam on half the other hands, it seems substantially better than blasting.

It matters here what "better than a finesse" and "worse than a finesse" and "no play" mean. No play might mean no play if they cash the CAK. I had a few where if they don't cash we are anywhere from a favorite to make 6, to cold for 13 tricks. There are a few that are 5/7 or 4/7 hands now. Again, the free parameter about how the opponents might lead. You have to put percentages on making and then compute the expectation over all the outcomes times the probability of each outcome to arrive at a net expectation for just bidding slam.

awm, on May 1 2008, 01:26 AM, said:

This suggests that blasting 6♠ may well lead to going down more than it leads to making. If we assume that the quantitative approach bids the "better half" of the slams you can see that it's a huge winner over blasting slam. More realistically, even if the quantitative approach just keeps you out of the "no play" slams and randomly gets you to slam on half the other hands, it seems substantially better than blasting.

Computing the expectation for bashing versus 2H...4S introduces a second free parameter, which is how often partner will move over the latter action by responder. I don't have a copy of bridge browser handy so I will just take a guess based on my experience that in real life it's between 30% and 40%. For computing the expectation of 2H... 4S we also need to add in the unbid slams when opener is passing and slam is still good.

I computed the relative expectation of the two bidding plans as a function of my simulation results and various values for the free parameters. I probably underestimated the chances of the opponents finding the correct lead when their choice matters, but I probably overestimated the chances partner would move over 4S. Reasonable people might disagree on the results and I didn't mean to present them as absolute. This all got away from my original point which is that the bidding problem is really just not a good bidding problem. I could have done a better job making myself clear in my original post and I apologize for that.

[ArtK78]

I admit I did not have the time (or the inclination) to read all of the foregoing posts carefully. But the idea that opposite a random strong 1NT opening that the 7222 hand would make 6♠ 80% of the time (or 65% of the time, or whatever number you finally wound up with) deserves a special comment (aside from the fact that I disagree with that assessment).

Is there some reason why we can't try to determine logically whether this is one of the 20% of the hands that will not make 6♠?

In addition, much of the discussion reminded me of a tournament recap that I am sure that I mentioned once or twice previously. In one of Edgar Kaplan's Bridge World recaps of a major tournament (it may have been a U.S. team trials) he commented about one hand in this manner: "Expert A appears to have doubled the opponents' 5♠ bid on the theory that on any given hand it is unlikely that the opponents can make 11 tricks in spades. Unluckily for Expert A, this was one of those hands."

[hrothgar]

xcurt, on May 1 2008, 08:41 PM, said:

awm, on May 1 2008, 01:26 AM, said:

I find this somewhat hard to believe. Surely with only 28 hcp between you or so, you could easily be off two aces or an ace and the trump king, or the club ace-king, etc.

This is measurable. I measured it. Partner has at least three of the CA, CK, SA, SK 85% of the time. That's from a simulation with a million hands generated including roughly 40,000 where partner has the strong NT.

Once again, I'd like to request that people who claim results based on simulations provide a copy of their code so that we can evaluate whether they know what they are talking about.

For example, here is a very simple script designed to measure Foo's claim

predeal

north SQJ87543, HAT, DAJ, C32

controls = hascard(south, AC) + hascard(south,KC) + hascard(south, AS) + hascard(south, KS)

condition

shape(south, any 4432, any 5332, any 4333) and
hcp(south) >= 15 and
hcp(south) <= 17

action

frequency (controls, 0, 4)

I readily admit that the definition a 1NT opening by South is quite loose. However, the main change that I'd add would be upgrading good 14 counts which would only decrease the control count.

Moreover, I haven't considered the opponent's pass which would modify the results (slightly). None-the-less, I find the results illuminating

Controls
0 Controls = 0%
1 Control = 2.9%
2 Controls = 41%
3 Controls = 50.1%
4 Controls = 5.9%

My sim says that Partner has a least three of the key controls about 56% of the time. There seems to be some inconsistency between our results.

Notice the convenience of being able to verify my methods...
Some would claim that this effects the credibility of my statement

[jdonn]

xcurt, on May 1 2008, 12:41 PM, said:

My simulation suggests that if we bash slam we will make it 65% of the time assuming the opponents lead randomly from among the offsuits.  That introduces a free parameter -- how good are the opponents on lead.  This parameter has a dependence on our auction.  The more information we transmit, the better they do.  The 65% figure is across all hands.

Wouldn't this strongly support making a slam TRY, as if partner accepts on about half of his hands you will make slam 100% of the time you bid it and 30% of the time you don't? I realize the correlation isn't absolutely as direct as that between partner's judgment and slam making, but it's obviously very strong so the conclusion should be the same.

[xcurt]

hrothgar, on May 1 2008, 04:58 PM, said:

...

You're correct and again I apologize for not being precise enough in my previous posts. I should have said "partner has two or more key cards," not "partner has 2 or more key cards and 3 of the 5 key cards and CK," at least 85% of the time. You also make a good point about code so here is the code to check. I'll try to post code for simulations when I do them.

# enough keys => we have 4 or 5 key cards between the NS hands
# clubs controlled => partner has at least one of the CA or the CK
# slam biddable => both of (enough keys) and (clubs controlled)

predeal

north SQJ87543, HAT, DAJ, C32

keys = hascard(south, AS)+hascard(south, KS)+hascard(south,AC)
enough_keys = keys>=2
clubs_controlled = hascard(south, AC) || hascard(south, KC)
slam_biddable = enough_keys && clubs_controlled

condition

shape(south, any 4432, any 5332, any 4333) and
hcp(south) >= 15 and
hcp(south) <= 17

action

average "enough keys" enough_keys,
average "clubs controlled" clubs_controlled,
average "slam biddable" slam_biddable

enough keys: 0.846463
clubs controlled: 0.882205slam biddable: 0.751074
Generated 1000000 hands, produced 33295.

But, we will make slam a fair bit of the time when clubs are open if the opponents lead poorly. So I'll stand by my conclusion that the overall expectancy of bashing is higher than 2H... 4S.

[xcurt]

jdonn, on May 1 2008, 05:42 PM, said:

xcurt, on May 1 2008, 12:41 PM, said:

My simulation suggests that if we bash slam we will make it 65% of the time assuming the opponents lead randomly from among the offsuits.  That introduces a free parameter -- how good are the opponents on lead.  This parameter has a dependence on our auction.  The more information we transmit, the better they do.  The 65% figure is across all hands.

Wouldn't this strongly support making a slam TRY, as if partner accepts on about half of his hands you will make slam 100% of the time you bid it and 30% of the time you don't? I realize the correlation isn't absolutely as direct as that between partner's judgment and slam making, but it's obviously very strong so the conclusion should be the same.

Yes but which slam try? The point I'm arguing, poorly (and that's good reason not to post at 12:30 am local or during lunch at the office), is that we should never have bid 2H... 4S since the simple bashing auction is already better expectancy.

Which bidding plan would you pick over a strong NT?

(a) sign off in game
(b) 2H...4S
© 2H...3C
(d) 2H...3D
(e) 2H-2S-6S
(f) 6S
(g) Stayman (followed by what?)
(h) other (what?)

We already know (a) is hopeless. I'm arguing that (e) is better than (b).* I expect (e) is better than (f) but I haven't tested it. (g) seems hopeless since we don't have a way to force in spades now and the information we get won't help. I have a sneaking admiration for plan © but I would probably choose (d) at the table.

*Unless partner is a madman who makes a forward-going move 65% of the time.

[awm]

My simulations indicate that signing off in game is probably better than signing off in slam. If anyone really wants to see my C code, I can send it to them -- I am not using the same dealing software that seems to be the rage these days. Certainly if opponents always make the right leads, it is not even close and signing off in game is dramatically better than signing off in slam.

While there are always opportunities for opponents to make truly awful leads, in most cases a passive lead against 6♠ doesn't cost on these hands. If we assume that opponents lead from small cards where possible (i.e. don't lead away from their club king into declarer's AQ or bang down the spade ace felling partner's singleton king) then signing off in game still looks like a winner over signing off in slam, albeit by a bit less.

Anyways, regardless of what the "right" answer might be, isn't the fact that this hand generates so much debate an indication that it was in fact a good problem hand?