correct values, falsified statistics and an apology

A few days ago I posted here the question, if some other Software is calculating correct or if only mine, because I calculated for hard 12 against a 4 with a well-balanced 1 deck left in the shoe the better advise is to hit.
While discussion we came to the point were we compared our expected values for that situation. Aside from a bug in my ! Software concerning H17, after I fixed it we all (MathProf, Cacarulo, a third person who emailed me and I) have had the same expected values for that situation:
(S17) Hit:-20.71% Stand:-21.36%
(H17) Hit:-20.74% Stand:-20.81%
These EVs indicate, that you should hit in the discribed situation, but I was told that this situation is simply irrelevant.
I wouldn't say this so hard. I think it is not a bad assumption that the remaining cards in the shoe are well-balanced if you have no information or no more than the visible cards. Composition dependent BS delivers the correct decision only in the first rounds, and is getting slightly worse. "Generic" BS too, but not as much. If you base your decisions on the quantity of the remaining cards in the shoe and assume that the shoe is well-balanced, your decisions will getting round by round slightly better. This could be very complicated and the useful effect at the end is very small.
It is not my goal to develop a new BS or to complicate the situation, it is my goal to develop a good software product that gives Black Jack players and Strategy-developers amaizing possibilities to analyse the game. Perhaps it is not yet ripe, but I am still in the introductory phase. If you download and test it you easiely recognize the big potential it has.
The discussion about hard 12 vs. 4 helped me to find the bug in Black Jack Advocate concerning H17, and I thank you for that. I apologize to have been so sceptical for your values, but beeing sceptical until a prove is simply my nature, especially concerning math. I belive only to statistics, that I falsified myself (This is translated and possibly bad interpreted from my german language, I hope you don't misunderstand it - of course I do not falsify statistics!!!). After the bugfixing I calculated the same values as you, that was the prove I needed. After that I compared a sample of expected values with http://www.bjmath.com/bjmath/ev/ev.htm , and except splitting decisions my values correspond exact. Due to the multitude of cards that must be considered in splitting decisions I suppose that Cacarulo's values on that Site for splitting decisions are estimations as well as mine, and that we only used different approaches. The differences between my values and Cacarulo's concerning splitting decisions are very small, I didn't find one that exceeds 1% difference.
If you are interested in Black Jack Advocate, please follow the Link to the Black Jack Advocate Homepage. In the latest version the H17-bug is fixed, as well as a "DLL missing"-Error. You will find an update on my Site, too.

It takes a big man and a true gentleman to admit he has made a mistake. I am interested in your product when you can input a +/- count and calculate on a variable amount of cards left. Please keep us posted on your progress and don't be a stranger to these pages.

Thank you.
I plan to integrate a count input and count analysing methods soon.
But first I enhance Black Jack Advocate to handle correctly the cases, that Roger R. Baldwin called in his "The optimum strategy in Blackjack" the "pathological" cases. They are not so rare and the correct calculation of the best advise what to do in these cases differs from that in normal cases. I will explain this more detailed when ready.

There is no need to apologize for doubting us. I am glad you found your bug.

These BJ computations are very complex, and it easy to have bugs in the program. When you first mentioned that your S17 figures were the same as mine, but the H17 were different, I wondered if I had made an error. I tended to doubt that, because I have been using these functions for a long time. But I double checked the "main" program to see if I had put in a parameter wrong. It wasn't until Cac said he got the same thing that I did that I became very confident that we were both right.

To get back to your question about other software. There is a program out called "Blackjack Subsets". I am not familiar with it. But I believe it does combinatorial calculations as well, and it may do the same things you are asking about. Maybe someone who has a copy can comment on it.

You also mentioned that you got different split values than Cac. Splitting is very complex. In fact, my code for splitting is probably more than all the other plays combined.

There is an algorithm in Griffin's book for splitting. It is exact for single deck (except for 10-splits). It doesn't quite handle re-splitting properly for multiple deck play. Steve Jacobs was the first person I know of who mentioned that.

Now I have a question for Cac: Did you use Griffin's algorithm? Or something else.

BTW, an even harder problem is computing the standard deviation for splitting. I have a good way of approximating it, which I posted once on BJ-Math. I am not aware of it being published anywhere.

and your program looks very user-friendly, if a bit short on rule variations. I'm looking forward to trying it out.

Now I have a question for Cac: Did you use Griffin's algorithm? Or something else.

I use something else. Here's a link to a related post where I explain the method I use:

BTW, an even harder problem is computing the standard deviation for splitting. I have a good way of approximating it, which I posted once on BJ-Math. I am not aware of it being published anywhere.

Agree. It's a very hard problem.

Sincerely,
Cacarulo

I am very happy to hear, that you think my software looks user friendly, because I think this is one of the most important parts of a good software.
I hope your test confirm your first impression, and if not or if you have some wishes, please inform me.
It's right that it is a bit short in rule variations. I will change this after I integrated counts. In the meantime you must unfortunately work around for some rules. Example: Surrender is not included yet, but you can check it via the Expected Value: If that is lower than -50% then surrender, else do what Black Jack Advocate advises.

Actually, Griffin's method isn't exact even for the single deck case.
It might be exact for any number of decks in the special case where
only a single split is allowed, but even that case might be an approximation.

I am aware of only flaw in Griffin's method, and that is the way it deals with re-splitting That flaw does not effect splitting where re-splitting is not allowed. It also doesn't effect splitting non-tens in single deck.

Perhaps there is another flaw that I am not aware of. Would you be able to elaborate.

I realize that this board is not condusive for high-level discussions. If you would rather post it somewhere else, like on one of Reid's boards, that would be fine.

Thanks.