[McBruce]

A BBO ACBL Virtual Club game, translated to ACBL Live for Clubs, link here

My question: who is the best C pair?

13 tables played 18 boards with a 12 top, so average is 6 * 18 : 108.

Pair 8 NS and 2 EW both have 111 matchpoints, and yes, I checked the individual boards.

Problem appears to be that Board 12, where there were two artificial averages, has a top of 10, not 12. ACBL is not using the formula to factor to the right top when they translate. This leads to standings where the same number of matchpoints shows as two different percentages.

We should thus use the BBO percentages, not the ACBL Live matchpoints.

(If it were my choice, I would certainly prefer that they post the boards here (except when the are Common Game hands that might be in play somewhere) before fixing this problem.)

[mycroft]

I can't find anything from ACBLscor->live4clubs with an average. Of course, I almost never give averages. Actually, I know there was one where I had to (web movement, one set of boards had 10's cards in both 10 and 11, and then it continued through to 14 or so), but I can't find it ATM. But my memory is that it factored those games correctly.

[McBruce]

mycroft, on 2020-May-07, 10:12, said:

I can't find anything from ACBLscor->live4clubs with an average. Of course, I almost never give averages. Actually, I know there was one where I had to (web movement, one set of boards had 10's cards in both 10 and 11, and then it continued through to 14 or so), but I can't find it ATM. But my memory is that it factored those games correctly.

I did provide a link....

[mycroft]

No, sorry. I saw your game, of course. What I was saying was that I could not see any of my RL games that had an AVE and whether it factored those right.

If it did - then it's something weird with the transition between BBO and Live4clubs. If it does the same thing (because I know ACBLScor does it right), then it's something wrong with Live.

Basically I was hoping to bisect the issue so we would know with whom the problem lies.

[barmar]

The scoring is being done by the BBO program that sends the results to L4C.

I wouldn't be surprised if our factoring algorithm for average boards is different from ACBLScor. When I was looking at our code, I found some of it dubious.

[mycroft]

To be clear, "Correct" = "Neuberg", at least according to ACBLScor (therefore, the ACBL). It looks like instead of that, we're taking out the relevant "averages" and then matchpointing to a N-Averages top. Which is fine when it will then become percentages (10/10 is still 100%), but not great when it's scored as matchpoints (as that board is suddenly worth 10/12th of the other ones).

My guess is that in BBO even for tournaments, everything is converted to percentages on the board and added-and-divided as percentages. In which case it wouldn't matter at all - 50% is still 50%, 60% is still 60%; the small differences in the fact that Neuberg would give that 10/10 on a 12 top something around 11.9 instead of 12 is - okay, wrong, but "almost" a wash. It's what people who hate Neuberg don't want ("but even if it's played another 17 times, nobody's going to get the +3400 we got!" Well, yes, there's a chance one of them would be just as mindless as your opponents, and play even worse!) so it might even be 'better" for player gripe versions of better.

(Yes, I remember a huge thread 6 or 7 years ago complaining "Why should my 100% beating 15 pairs not be worth more than your 100% beating 4?", but *usually* it's the other way around).

[McBruce]

Today's 0-300 game had a very amusing result, 3rd in C listed ahead of 2nd in C. Here is the link. This group had some problems getting finished on time, proving really proficient at hitting round ends in spots where key decisions were still to be made, and the sub pair was actually a quintet by the time it was over, so there were several artificial averages given. I am reminded of the infamous Murray Walker F1 quote: "the third-place car has just lapped the second-place car..."

[McBruce]

barmar, on 2020-May-07, 13:05, said:

The scoring is being done by the BBO program that sends the results to L4C.

I wouldn't be surprised if our factoring algorithm for average boards is different from ACBLScor. When I was looking at our code, I found some of it dubious.

To be perfectly clear, the ACBL conversion from what BBO sends to Live seems to be the problem here. It is reporting the BBO percentage scores for pairs, but then computing their matchpoints based on board results, without properly factoring boards where there are artificial averages. BBO may well have it right: the formula for Neuberg does look a bit dubious.