[inquiry]

Got in a discussion on what the minimum values for 3NT is these days. Goren, of course, suggest 26 hcp. So I thought a short poll to see what the forum members consider "standard" count (plus other values if you care to want to insist on something) is for bidding to 3NT with a balanced hand opposite a balanced hand.

Background, those who play Kaplain and sheiwold know that tehy beleive balanced 12 opposite balance 12 is enough for game. So that is 24. So if you want to add some stipulations, like if the hcp are or are not evenly divided, feel free to add comments to this thread. For instance, if one hand had 20, would 6 be enough for opposite? would 5 be? Things like that. IF you upgrade combinations of honors, or multiple 10's or "good intermediates" (lets say 9's), use the multiple "10;'s" in the poll.

[mikeh]

Doesn't it depend on vulnerability and scoring method?

I think at mps, against strong opps, the breakeven point would be almost the playing strength of a full point more than at imps, when red. At imps, red, I like 24 with a 5 card suit or a couple of 10's. NV, I play it essentially the same as at mps. So any 25, for sure at mps or imps nv... and great 24s.

[awm]

Well, I make a lot of upgrades and downgrades for various reasons.

Assuming "nothing special" about the hands my guess is that 3NT will make about half the time on 25 hcp, a bit less (but not horrible) on 24 hcp. This means at IMPs I always want to be in 3NT with 25 hcp, whereas 24 hcp without any special features is fine either way. At MPs I prefer to avoid 24 hcp 3NT (unless really exceptional hands) and don't mind missing a 25 hcp 3NT on very unexceptional hands.

So for example holding a balanced hand opposite 15-17 (and a partner for whom 15-17 doesn't mean "a really nice 13 to a mediocre 16") I will game force all ten-counts. At MPs I will pass most eight-counts (only inviting on the truly exceptional ones) and will invite most nine-counts (GF on exceptional ones). At IMPs I will game force most (but not all) nine-counts and will invite on most eight-counts.

[MrAce]

I am talking about imps. I am from Europe and MP is not even considered as serious bridge there I don't really pay much attention to our combined hcps, but spot cards.

When we bid a game, a good outcome depends on 4 things;

-What we have

-The Lead

-Defense

-Our opponent's shape

Unfortunately our combined hcp, is the LEAST important of four elements that decides the good outcome in a borderline game imo. Arriving to the game with least amount of info to opponents, increases the chance of making borderline games.(Avoiding the invitation bids as much as we can) It directly effects THE LEAD and DEFENSE. Having rich spots makes it extremely hard for defense to reach to 5 tricks.

So i chosed the minimum (24) hcp option with spots.(There was no option that says "24 hcp with rich spots and a fast bidder" I think combined 24 hcps with rich spots almost always have at least %40 chance when arrived fast to game.

One opens 1 NT, other one invites with 9 due to shyness, other one passes with 15 and accepts if he had an extra J. This is not even scientific as some top Italian players stated in the past.

[gwnn]

Just look at your hand?

but sure, all other things being equal, 25 should be 3NT and when I see 24 with at least one redeeming feature (such as the realistic possibility of having 25, or honour structure or a 5 card suit) I also want to be there

[nige1]

inquiry, on 2010-December-07, 22:48, said:

Got in a discussion on what the minimum values for 3NT is these days. Goren, of course, suggest 26 hcp. So I thought a short poll to see what the forum members consider "standard" count (plus other values if you care to want to insist on something) is for bidding to 3NT with a balanced hand opposite a balanced hand.
Background, those who play Kaplain and sheiwold know that tehy beleive balanced 12 opposite balance 12 is enough for game. So that is 24. So if you want to add some stipulations, like if the hcp are or are not evenly divided, feel free to add comments to this thread. For instance, if one hand had 20, would 6 be enough for opposite? would 5 be? Things like that. IF you upgrade combinations of honors, or multiple 10's or "good intermediates" (lets say 9's), use the multiple "10;'s" in the poll.

IMO It depends on scoring method, vulnerability, and how evenly the strength is split. At the extremes
Vulnerable at at imps, 12 opposite 12, is ample.
At match-pointed pairs, 25 opposite 1 is doubtful.

In practice, due to the vagaries of notrump ranges, you often declare 3N on 23. For example, opposite a 15-17 1N opener what do you reply with
♠ xx ♥ xx ♦ xxxx ♣ AKJxx

AFIR, there is an article about this on Richard Pavlicek's excellent site.

[gwnn]

i call the director (sorry!)

[Free]

The question is constructed very poorly. There's not a single situation based on HCP, tens and a possible 5 card suit that I'd always want to be in 3NT. Sometimes I'd want to be in 4M or 5m. Plus the vulnerability and scoring can be very important (imps V is not the same as MP V)

You can't just say with such and such hand I'd always want to be in 3NT. For example I prefer to play 3NT with 12 vs 12 rather than 23 vs 1. Still, in both cases we have 24HCP. Same goes for tens, intermediates, quacks,... What's the point in having a 5 card suit and 24HCP if you can never reach the 5 card suit (24 vs 0 with a 5 card)?

[lexlogan]

If I know we have 25 hcp, two aces, and either a ten or a five card suit, I want to be in 3NT at any form of scoring. This does not mean, however, that I want to risk 2NT when that's the best we might have. So I wasn't quite sure how to answer the poll.

[mikl_plkcc]

This is the question I always think about.

A deck has 40 HCPs, so it is approximately 3 HCP per trick, so 27 HCPs make 9 tricks, and 26 HCPs have a large likelihood to make 9 tricks (here, we assume that all 4 hands are 4333 and all honours are distributed evenly among the suits and among both hands).

Why is 25 HCPs the commonly-accepted values to bid 3NT?

[mw64ahw]

There may be some stats. on Pavilecks website, but I seem to remember that 25 was ~53% and 24 ~48% likelihood of making. I think this is also in line with my own sims.

[bluenikki]

inquiry, on 2010-December-07, 22:48, said:

Background, those who play Kaplain and sheiwold know that tehy beleive balanced 12 opposite balance 12 is enough for game. So that is 24.

12 facing 12 yes. 20 facing 4 no. 16 facing 8 no. 15 facing 9: most seem to say yes, but too often the 9 won't have enough entries.

[bluenikki]

mikl_plkcc, on 2025-February-25, 09:49, said:

This is the question I always think about.

A deck has 40 HCPs, so it is approximately 3 HCP per trick, so 27 HCPs make 9 tricks, and 26 HCPs have a large likelihood to make 9 tricks (here, we assume that all 4 hands are 4333 and all honours are distributed evenly among the suits and among both hands).

Why is 25 HCPs the commonly-accepted values to bid 3NT?

Because 4333 is rare.

[mikl_plkcc]

bluenikki, on 2025-February-25, 14:11, said:

Because 4333 is rare.

So if there is some distribution, the opponents also have some distribution as well. If we can make a distributional 3NT with less HCP, there is also a balancing chance that the opponents can run off a suit even if we have more HCP. So where is my math goes wrong?

[awm]

mikl_plkcc, on 2025-February-25, 09:49, said:

This is the question I always think about.

A deck has 40 HCPs, so it is approximately 3 HCP per trick, so 27 HCPs make 9 tricks, and 26 HCPs have a large likelihood to make 9 tricks (here, we assume that all 4 hands are 4333 and all honours are distributed evenly among the suits and among both hands).

Why is 25 HCPs the commonly-accepted values to bid 3NT?

There is some combining effect of high card points because honours work together. You can see this most clearly in the high card numbers for slam: 33 high card points is 82.5% of the total but you expect to make 92.3% of the tricks. It also works the other way, if your side has two aces and no other high cards (and no distribution to speak of), you will take two tricks (15.4% of the total tricks) even though you have 20% of the total high card points (and most analyses have aces being undervalued by the standard point count). To see high cards working together, if you have an ace and a king in the same suit you will likely take two tricks (again, assuming no weird distribution) whereas an ace and a king in different suits are only around 50% to take two tricks (depending on whether the ace is in front of or behind the king).

Anyway, the high cards for game are determined based on some combination of double dummy analysis and real play experience (they are a little different; in particular declarer has a bit of an advantage in real play mostly because of the "blind" opening lead before dummy is faced).

[pescetom]

 mw64ahw, on 2025-February-25, 10:02, said:

There may be some stats. on Pavilecks website, but I seem to remember that 25 was ~53% and 24 ~48% likelihood of making. I think this is also in line with my own sims.

I seem to remember less than that, for both. But I was using double dummy tricks and assumed a somewhat liberal strong 1NT opening.

[helene_t]

Csaba (gwnn) once made various gib vs gib matches to see what linear combinations of honours predict what contracts can be made. For 3NT, his coefficients were
A 4.076
K 2.931
Q 1.983
J 1.010
T 0.588

So almost identical to Walrus points. I made a similar analysis of GIB's DD database and found slightly different results, in particular the value of the queen was much lower which I thought might be due to a queen being less valuable DD when declarer always gets the two-way finessed right (and opps know when not to underlead AKxxx and such).

In the Vugraph archive there are just short of 300000 3NT contracts, of which 70.8% made. So if you guess "made" always, you get it right 70.8% of the time. Using WalrusPoints>24 as predictor, you guess right 72.0% of the time.

Using simple single-honour coefficients like in Csaba's model at p(tricks>=9)>0.5 as cut-off, you guess right 73.7% of the time. Those coefffcients (rescaled to A+K+Q+J=10 to make them consistent with Csaba's way of reporting) are
A 4.135323
K 2.924272
Q 1.923956
J 1.016449
T 0.530461

so human play values aces a bit more than Gib play does, but they are very similar.

Taking all possible two-way interactions between honours in the same suit, plus suit length main effect as a categorical predictor and suit length honour interactions based on linear suit length, you get a model that can guess right 74.4% of the time. The model becomes too complex for me to interpret but something that is maybe of a bit of interest is the contribution to the deviance from honour:suitlength interactions:
A 44
K 112
Q 198
J 166
T 90

which is maybe a bit surprising, it looks like it doesn't matter so much in which suit your aces are, compared to the other honours.

This is still not the perfect model, of course. Linear suit length is quite crude an also, synergy between declarer's and dummy's holding in the same suit could be interesting.

I have implemented some of Csaba's models as a web app so you can play with them here: https://helene-h-thy...and-evaluation/

[helene_t]

BTW, the different studies use different sampling frames. The Vugraph Archive is hands where people actually bid 3NT. Csaba's analysis and my DD analysis were based on some plausable-3NT criteria, I don't remember the details.

We would probably expect the models to give better predictions on synthetic data than on human play. This is because when humans bid a 23-point 3NT there are likely to be some compensating factors which I don't take into account, such as a finesse likely to work based on the bidding.

Maybe this doesn't affect the relative size of the coeffcients. But it could explain why aces are more valuable in human play than in GIB play - aces are maybe less affected by opps' bidding.

I think the best way around this would be to run the Gib-vs-Gib and DD analyses on the Vugraph data. But that would just mean that we impose the same bias on the sims as we do on the real data. Alternatively, I could take information from the auction into account in the model, or maybe adjust for whether opps found the DD lead.

[awm]

helene_t, on 2025-March-09, 06:28, said:

BTW, the different studies use different sampling frames. The Vugraph Archive is hands where people actually bid 3NT. Csaba's analysis and my DD analysis were based on some plausable-3NT criteria, I don't remember the details.

We would probably expect the models to give better predictions on synthetic data than on human play. This is because when humans bid a 23-point 3NT there are likely to be some compensating factors which I don't take into account, such as a finesse likely to work based on the bidding.

Maybe this doesn't affect the relative size of the coeffcients. But it could explain why aces are more valuable in human play than in GIB play - aces are maybe less affected by opps' bidding.

I think the best way around this would be to run the Gib-vs-Gib and DD analyses on the Vugraph data. But that would just mean that we impose the same bias on the sims as we do on the real data. Alternatively, I could take information from the auction into account in the model, or maybe adjust for whether opps found the DD lead.

This brings to mind a study that I'm very curious about. A while back, a pair of books was written by Bird and Anthias, which analysed the effectiveness of various leads on a double-dummy basis. The main complaint about these books was that there are differences between double-dummy and real life declarer play. The analysis I'm interested in would look something like this (all hands from Vugraph archive):

1. How often does declarer make the contract when opponents start with a double-dummy setting lead?
2. How often does declarer make the contract when a double-dummy setting lead exists, but opponents did not find one?

The hypothesis would be that 1 >> 2; if this is not true it would tend to invalidate the Bird-Anthias assumption that you should try to find the double-dummy setting lead. Perhaps more on-topic though:

3. If opponents make a passive lead (from only small cards against a notrump contract, or from 3+ small cards against a suit contract), how does declarer fare vs. double-dummy after the lead?
4. If opponents make an active lead (away from one or more honors, not from an honor sequence), how does declarer fare vs. double-dummy after the lead?

The claim of most people objecting to Bird-Anthias is that 3 >> 4 for declarer. This seems somewhat believable because the active lead often either sets the contract or lets it make without a lot more going on, whereas the passive lead can require very accurate decision-making by defenders later in the hand. Against this, the active lead sometimes "gives a guess" which declarer would always get double-dummy anyway (although this can also happen after a passive lead if partner has queen in the suit).

[helene_t]

I am about to do such an opening lead study but I am not sure where to publish it. I might try Bridge World and if they don't want it I will just post it here.

I haven’t incorporated DD though, I am struggling a bit that, need to learn Python.