Hands played at the Palmerston North Bridge Club published here for discussion and question and answer.
Any bidding shown is an actual auction and not necessarily recommended.
Please feel free to participate and make suggestions.
A combined 32hcp balanced with pretty low chance of making 12 tricks. Should east only invite with his hand? DAA says the chance of success is 67% opposite a 20hcp, which, allowing for DDA bias in slams seems OK. Testing for generic 12hcp hands with 4=4=(3-2) with no major fit, against 20hcp balanced is 73%. There are two negative features of the east hand,xxxx, and no tens. Testing for each of these gave prob success of 63% and 67%, ie both are negative features, which combined probability lead to a marginal decision. Unfortunately the west hand in min with negative features (4333), and no tens.
I am surprised the numbers are that high. However I got similar numbers 64% opposite any 20 count and 71% opposite a hand with no major.
However I then looked at 20 hands where it said I could make 6NT and I reckoned on 5-6 of them I would not or may not make double dummy. One I could not see the double dummy line to make after about 5 minutes - must have been a squeeze. I will put it into a double dummy engine and check. And another one where there was a two way finesse with not much to go on so you might go wrong. But there were four where normal play failed and you had to perform miracles to make. On the other hand there was one hand I noticed where single dummy was easier than double dummy to get twelve tricks. Double dummy a heart lead from Qx was best - i figured this was not a likely lead. Then declarer had to get a two way finesse right to make. Without the heart lead declarer had enough tricks without getting the two way finesse right.
So that makes around 25% of the time you fail when double dummy says you would succeed. There is probably some flip side where double dummy says you will fail but single dummy you might make but I suspect those are much less frequent.
That means my 71% reduces to about .75 * 71% = 52%. So it is getting very close to a line-ball 6NT. On the inconclusive evidence gathered so far slightly in favour of bidding 6NT.
Ok, it looks like my early work indicating DDA bias on slam hands suggesting on average about a 8% bias may have been faulty. May still be useful when checking for relative negative features.
Also, what do u think about cranking up SDA to compare what the robots achieve vs DDA on generic balanced NT slams? that might give some idea of the relative bias?
Also could use SDA as a direct test of the actual hand problem with say 100 hands?
I cranked up SDA for 100 hands, but the results I report are for 50 hands (I can do the analysis on the further 50 on request). I used a combined 32hcp with only 4432 and 4333 hands (for both) and 15-17 hcp in one hand (GIB opens 1NT with this). I ran DDA on the same hands.
DDA was 74% made 12+, SDA was 70% made 12+. GIB bid slam on 66%, but didn't seem to have a super system as on 10 hands they didn't bid but made slam, while 7 hands they didn't bid and didnt make, with a further 6 where they bid slam but didn't make.
There were 6 hands where DDA said 6 could be made but wasn't by GIB (ie DDA bias of ~12%), and 2 hands where DDA said 6 couldn't be made but was (ie probably opening lead counterbias of 4%). There were a further 4 hands (8%) where DDA said it could make 7 but GIB made only 6.
Not 100% sure but I think there is another bias in SDA if GIB did not bid six. In that the defenders will or may be trying to defeat three no trumps and in the process give away overtricks.
I think you should be able to specify the auction to 6NT and then just let GIB play the hand.
I reanalysed the existing 100 hands into two groups, bid slam, didn't bid slam. The net DDA bias of the bid slam group (60 hands) was 10% (75% vs 65%), whereas for the didn't bid slam group it was 15% (55% vs 40%). Not sure whether that is just noise difference or supporting your theory.
I analysed the full 100 hands. The "net" DDA bias was 12%, with DDA getting 68%, SDA 56%. 3 (3%) hands were where SDA beat DDA, so eliminating those as "opening lead" bias gives u 15% gross DDA bias. Some finer grain analysis revealed that the DDA average trick bias was greater than the "make contract bias" as seven of the hands where neither DDA nor SDA made 6, then DDA was >SDA, and also five hands where DDA made 7 but SDA 6.
I have now extended the same analysis to the game situation with 25hcp combined. DDA was 51% and SDA was 52%. Not same 50+% of course, with DDA making/SDA not making ~11% but SDA making/DDA not making 13%. GIB bid to game 78% of the time, and in fact its likelihood of making game was slightly better if it had not bid it! Curiously DDA was better when both DDA/SDA <game and SDA was better when both were game+.
Compared with the slam situation there is gross DDA bias of a similar level in both, but at game level this is entirely compensated by SDA bias.
A combined 32hcp balanced with pretty low chance of making 12 tricks. Should east only invite with his hand? DAA says the chance of success is 67% opposite a 20hcp, which, allowing for DDA bias in slams seems OK. Testing for generic 12hcp hands with 4=4=(3-2) with no major fit, against 20hcp balanced is 73%. There are two negative features of the east hand,xxxx, and no tens. Testing for each of these gave prob success of 63% and 67%, ie both are negative features, which combined probability lead to a marginal decision. Unfortunately the west hand in min with negative features (4333), and no tens.
ReplyDeleteI am surprised the numbers are that high. However I got similar numbers 64% opposite any 20 count and 71% opposite a hand with no major.
DeleteHowever I then looked at 20 hands where it said I could make 6NT and I reckoned on 5-6 of them I would not or may not make double dummy. One I could not see the double dummy line to make after about 5 minutes - must have been a squeeze. I will put it into a double dummy engine and check. And another one where there was a two way finesse with not much to go on so you might go wrong. But there were four where normal play failed and you had to perform miracles to make. On the other hand there was one hand I noticed where single dummy was easier than double dummy to get twelve tricks. Double dummy a heart lead from Qx was best - i figured this was not a likely lead. Then declarer had to get a two way finesse right to make. Without the heart lead declarer had enough tricks without getting the two way finesse right.
So that makes around 25% of the time you fail when double dummy says you would succeed. There is probably some flip side where double dummy says you will fail but single dummy you might make but I suspect those are much less frequent.
That means my 71% reduces to about .75 * 71% = 52%. So it is getting very close to a line-ball 6NT. On the inconclusive evidence gathered so far slightly in favour of bidding 6NT.
Ok, it looks like my early work indicating DDA bias on slam hands suggesting on average about a 8% bias may have been faulty. May still be useful when checking for relative negative features.
ReplyDeleteAlso, what do u think about cranking up SDA to compare what the robots achieve vs DDA on generic balanced NT slams? that might give some idea of the relative bias?
Also could use SDA as a direct test of the actual hand problem with say 100 hands?
I cranked up SDA for 100 hands, but the results I report are for 50 hands (I can do the analysis on the further 50 on request). I used a combined 32hcp with only 4432 and 4333 hands (for both) and 15-17 hcp in one hand (GIB opens 1NT with this). I ran DDA on the same hands.
ReplyDeleteDDA was 74% made 12+, SDA was 70% made 12+. GIB bid slam on 66%, but didn't seem to have a super system as on 10 hands they didn't bid but made slam, while 7 hands they didn't bid and didnt make, with a further 6 where they bid slam but didn't make.
There were 6 hands where DDA said 6 could be made but wasn't by GIB (ie DDA bias of ~12%), and 2 hands where DDA said 6 couldn't be made but was (ie probably opening lead counterbias of 4%). There were a further 4 hands (8%) where DDA said it could make 7 but GIB made only 6.
Not 100% sure but I think there is another bias in SDA if GIB did not bid six. In that the defenders will or may be trying to defeat three no trumps and in the process give away overtricks.
DeleteI think you should be able to specify the auction to 6NT and then just let GIB play the hand.
I reanalysed the existing 100 hands into two groups, bid slam, didn't bid slam. The net DDA bias of the bid slam group (60 hands) was 10% (75% vs 65%), whereas for the didn't bid slam group it was 15% (55% vs 40%). Not sure whether that is just noise difference or supporting your theory.
DeleteI analysed the full 100 hands. The "net" DDA bias was 12%, with DDA getting 68%, SDA 56%. 3 (3%) hands were where SDA beat DDA, so eliminating those as "opening lead" bias gives u 15% gross DDA bias. Some finer grain analysis revealed that the DDA average trick bias was greater than the "make contract bias" as seven of the hands where neither DDA nor SDA made 6, then DDA was >SDA, and also five hands where DDA made 7 but SDA 6.
ReplyDeleteI have now extended the same analysis to the game situation with 25hcp combined. DDA was 51% and SDA was 52%. Not same 50+% of course, with DDA making/SDA not making ~11% but SDA making/DDA not making 13%. GIB bid to game 78% of the time, and in fact its likelihood of making game was slightly better if it had not bid it! Curiously DDA was better when both DDA/SDA <game and SDA was better when both were game+.
ReplyDeleteCompared with the slam situation there is gross DDA bias of a similar level in both, but at game level this is entirely compensated by SDA bias.