[thepossum]

Hi

Apologies if these are covered in historic threads. I have tried searching and can't find anything very recent on the matter

It relates to some of the core assumptions, terminology and discussions about DD. Excuse me if my questions seem naive

1. Best Declarer vs Best Defence

To me that suggests if you had two top Bridge pairs playing each other, how they would play the hand - sorry it's like two expert declarers
Double Dummy on the other hand treats Best as knowing more about the layout than even an expert would deduce

2. Adjusting play for safety

Do many DD solvers make adjustments for safety when necessary
A simplistic statistical analysis would make errors - do the best ones always place making a contract ahead of "best" average score
Maybe (some) top players always go for best average

Hope those questions make sense. I have other similar ones but they are two that keep cropping up when I look at the solutions DD analysis provides

[steve2005]

DD declarer tries to maximize the number of tricks, knowing the layout of the hands.
It doesn't play safe as it knows if the finesse will fail.
The double dummy examines all possible plays and determines the best play given that the defender will also make the best play at it's turn.

[thepossum]

steve2005, on 2025-January-13, 21:42, said:

DD declarer tries to maximize the number of tricks, knowing the layout of the hands.
It doesn't play safe as it knows if the finesse will fail.
The double dummy examines all possible plays and determines the best play given that the defender will also make the best play at it's turn.

I know all that. I am asking more philosophical questions about what best declarer and best defence is

That is what I prefer to compare myself against

I believe some have even analysed slams and asked questions why professionals and other top players don't make (or even bid) many DD slams

Would it not be a cool mode to switch on and off. Professional or expert mode

[mycroft]

I know this has changed somewhat in the last 10 years, but (from a side direction) here's the issue:

Why are robots bad? Because they use double dummy simulations to "simulate" single dummy analysis.

Why do robots do this? Because single dummy analysis is an unsolved problem (or, at least, SDAnalyzers are currently worse than simulate from multiple DD analyses.

Given that, the answer to your question is "without paying an expert to analyze every hand SD, from both directions, there is no way to give you this answer."

We would all like to have "actual best bridge" analysis on our hands - especially those of us who are directors (or teachers) who have to explain (again) "we beat par by a lot! Why did we score 30%?" And yeah, I had one Sunday night, where "okay, 3NT goes down. But it requires the opening leader to *underlead* ♥AKQxxx at trick 1, so that partner still has a heart to lead when he gets in. I'll applaud the expert who finds that play."

TL, DR; you get DD because what you (and all of us) want Does Not Exist.

Take heart, though; it's also why chess cheaters get caught by "making multiple Stockfish best moves" after (sometimes deliberately, sometimes just their actual skill) playing to get behind in the game; and GIB playing with GIB in an average club game gets 54% for third or fourth.

[mikeh]

As stated earlier in this thread, DD analysis is based on examining all 52 cards….the layout of which is known with 100% accuracy.

Say one is playing in 4S and the key is the location of the spade king…dummy has AQ9x, we hold J10xxx, and we know that RHO has length in both minors and presumably short spades. With no other clue from the bidding or play, all good players will finesse against the king. But DD play always drops the king when it’s singleton offside and always finesses when it’s onside. No non-cheating good player would ever drop the king when it’s stiff offside (unless desperate for a swing and willing to look silly while doing so).

Do NOT ever confuse DD play with ‘expert’ or ‘professional’ play.not only do no players…(other than self-kibitzing cheaters) play double dummy but no good player would ever even try.

There are hands on which a skilled declarer has a very good idea about the opposing hands….I’ve held a few where I was morally certain of how the cards lay as early as trick one or two….but even then I’d not usually be certain about where, say, a missing jack might be. Such hands will occasionally arise when the opps have provided high card and distributional information during the bidding and then further inform from the opening lead and the play by third hand. In those situations it is possible to adopt a line that, to an observer naive to how much information declarer has been given, will appear to be a double dummy line. So there are times when, due to skill and the willingness to draw inferences, the play is basically DD

The key lies in how an expert player deduces the layout and then adopts a line of play. Computers don’t draw inferences. Some day AI may be able to do that but for now they analyze based on being given the data about exactly where every card lies. They don’t ‘care’ whether RHO showed, say, 12 cards in the minors….if the non minor card is the king behind dummy’s AQJx, they drop the king, not because they drew an inference ir played for a very low percentage swing but because they ‘know’ the king is offside.

I used to give lessons at a local club, using the previous week’s hand records. The GIB DD analysis made the effort largely wasted. The players who attended were often fixated on the DD analysis. ‘It says we can make 4S, but we didn’t bid it. What did we do wrong?’

My answer: Spades broke 3-1 with the king singleton offside. Hearts were 4-2, you had Kxx in dummy and AJxx in hand, you can find out (by playing other suits) that LHO has two and RHO had 4….You had to drop the doubleton Queen. So to make, you have to play very badly.

[pescetom]

I agree completely except for:

mikeh, on 2025-January-14, 12:40, said:

The key lies in how an expert player deduces the layout and then adopts a line of play.
Computers don’t draw inferences.
Some day AI may be able to do that

My view here is that current state of the art AI is quite capable of drawing inferences but has not been effectively applied to the problem of playing Bridge, unlike the problems of playing Chess or Go.
Probably because there is too little public interest and economic return, rather than intrinsic difficulty of playing bridge or the complication that is a pairs game with disclosure of agreements (current robots wouldn't even beat the humans at Whist).