Serious Topic: Legitimate Proposal for Worlds 2005

[NoPoke]

After 9 rounds players 19, 23, 25, 28, 29, and 31 don't make the cut with [Scrub??] 68 being the lowest ranked player who made the cut. Notice that both the players rated 23rd and 25th at the start of the tournament have now been ejected from the cut. They were in the cut after 7 rounds and now they aren't. Let the complaints that the longer tournament screwed them over commence LOL



I went to ten rounds including fighting TMS which insists on messing up the pairings since round 7. fortunately it was only messing up the very worst player so wasn't too hard to fix.

after 10 rounds the following players still miss the cut :-

19th
28th
29th
31st

The worst player who made the cut was ranked 49th

I suspect that it will take a very very large number of rounds before the TOP32 has settled to just the best 32 at the begining of the tournament. Sadly as the number of rounds increases TMS just gets worse and worse. At least after 7 rounds the best 16 entrants will make the cut. So my belief that the swiss is primarily to eliminate the weakest and let in a small number of wild cards seems justified. The common sense and obvious approach of changing the swiss to match play and more rounds just doesn't work.

Rankings (Round10)
Rank   Player                    Score     Wins      Losses    Ties      TieBrk
1      A A1                      30        10        0         0         73%
2      A A2                      27        9         1         0         72%
3      A A7                      24        8         2         0         71%
3      A A3                      24        8         2         0         70%
3      A A6                      24        8         2         0         66%
3      A A5                      24        8         2         0         64%
7      A A4                      21        7         3         0         72%
7      B B13                     21        7         3         0         69%
7      B B11                     21        7         3         0         68%
7      B B10                     21        7         3         0         65%
7      A A8                      21        7         3         0         64%
7      B B9                      21        7         3         0         64%
7      B B14                     21        7         3         0         64%
7      C C20                     21        7         3         0         64%
7      C C21                     21        7         3         0         62%
7      C C17                     21        7         3         0         61%
7      B B12                     21        7         3         0         60%
7      C C24                     21        7         3         0         60%
7      B B16                     21        7         3         0         59%
7      C C35                     21        7         3         0         59%
7      C C30                     21        7         3         0         58%
7      D D36                     21        7         3         0         58%
23     B B15                     18        6         4         0         65%
23     C C18                     18        6         4         0         60%
23     C C27                     18        6         4         0         60%
23     C C25                     18        6         4         0         59%
23     C C23                     18        6         4         0         59%
23     C C26                     18        6         4         0         59%
23     D D49                     18        6         4         0         59%
23     D D46                     18        6         4         0         58%
23     C C22                     18        6         4         0         58%
23     C C32                     18        6         4         0         57%

The following players miss the cut...


23     C C33                     18        6         4         0         57%
23     E E68                     18        6         4         0         57%
23     C C34                     18        6         4         0         56%
23     D D39                     18        6         4         0         56%
23     C C19                     18        6         4         0         55%
23     C C28                     18        6         4         0         55%
23     D D41                     18        6         4         0         55%
23     D D48                     18        6         4         0         55%
23     C C31                     18        6         4         0         53%
23     D D43                     18        6         4         0         53%
23     D D42                     18        6         4         0         53%
23     D D54                     18        6         4         0         52%
23     C C29                     18        6         4         0         51%
23     D D51                     18        6         4         0         51%
23     E E77                     18        6         4         0         50%
48     D D40                     15        5         5         0         60%
48     D D44                     15        5         5         0         57%
48     E E71                     15        5         5         0         55%
48     E E65                     15        5         5         0         54%
48     D D55                     15        5         5         0         54%
48     D D37                     15        5         5         0         54%
48     D D50                     15        5         5         0         53%
48     D D38                     15        5         5         0         53%
48     D D56                     15        5         5         0         53%
48     D D63                     15        5         5         0         52%
48     E E67                     15        5         5         0         52%
48     E E78                     15        5         5         0         52%
48     D D62                     15        5         5         0         51%
48     D D45                     15        5         5         0         51%
48     D D58                     15        5         5         0         51%
48     D D59                     15        5         5         0         50%
48     E E72                     15        5         5         0         50%
48     E E73                     15        5         5         0         50%
48     D D47                     15        5         5         0         49%
48     E E83                     15        5         5         0         49%
48     E E94                     15        5         5         0         49%
48     D D60                     15        5         5         0         49%
48     E E75                     15        5         5         0         48%
48     E E80                     15        5         5         0         48%
48     D D61                     15        5         5         0         47%
48     E E79                     15        5         5         0         46%
48     E E95                     15        5         5         0         46%
48     E E106                    15        5         5         0         45%
48     E E84                     15        5         5         0         45%
48     D D53                     15        5         5         0         44%
48     D D64                     15        5         5         0         44%
48     E E86                     15        5         5         0         44%
48     E E91                     15        5         5         0         44%
48     E E66                     15        5         5         0         43%
82     D D57                     12        4         6         0         54%
82     E E69                     12        4         6         0         50%
82     E E102                    12        4         6         0         49%
82     D D52                     12        4         6         0         49%
82     E E76                     12        4         6         0         48%
82     E E74                     12        4         6         0         48%
82     E E88                     12        4         6         0         47%
82     E E70                     12        4         6         0         46%
82     E E90                     12        4         6         0         46%
82     E E85                     12        4         6         0         45%
82     E E96                     12        4         6         0         45%
82     E E81                     12        4         6         0         42%
82     E E93                     12        4         6         0         42%
82     E E103                    12        4         6         0         42%
82     E E104                    12        4         6         0         42%
82     E E89                     12        4         6         0         41%
82     E E99                     12        4         6         0         41%
82     E E98                     12        4         6         0         40%
82     E E87                     12        4         6         0         40%
82     E E100                    12        4         6         0         40%
82     E E108                    12        4         6         0         38%
82     E E109                    12        4         6         0         36%
82     E E101                    12        4         6         0         36%
82     E E115                    12        4         6         0         33%
106    E E82                     9         3         7         0         46%
106    E E110                    9         3         7         0         45%
106    E E97                     9         3         7         0         43%
106    E E107                    9         3         7         0         42%
106    E E113                    9         3         7         0         42%
106    E E92                     9         3         7         0         41%
106    E E105                    9         3         7         0         40%
106    E E112                    9         3         7         0         40%
106    E E117                    9         3         7         0         38%
106    E E119                    9         3         7         0         36%
106    E E116                    9         3         7         0         36%
106    E E114                    9         3         7         0         36%
106    E E111                    9         3         7         0         35%
106    E E121                    9         3         7         0         35%
106    E E120                    9         3         7         0         34%
106    E E118                    9         3         7         0         33%
106    E E125                    9         3         7         0         31%
123    E E122                    6         2         8         0         37%
123    E E123                    6         2         8         0         32%
123    E E127                    6         2         8         0         31%
123    E E124                    6         2         8         0         30%
123    E E126                    6         2         8         0         28%
128    E E128                    0         0         10        0         31%

 

[PokePop]

You know, there's something I've always wondered. What happens if each round of Swiss were a random matchup, just like the first round.
How would that affect the top 32 cut?

 

[NoPoke]

have you ever used TMS ? After about 7 rounds it seems completely random.!

I suspect that the problem you are proposing actually has a closed form solution.

Assume strongest player always wins.
P players
for each round the ith rated player has a win probability of (1-(i-1)/P)

now what I want is to establish a confidence limit on the 32nd ranked player making the cut after R rounds... hmmm ... this may take me a little while. Or in other words HELP!

------------------------------------------------

I believe that the following is true but I'm not certain: After 7 rounds of random pairings with 128 entrants , approx 1 in every 7 tournaments you will have more than 32 players with a perfect record.

 

[Articjedi]

There is a massive problem with the experiment though.
2/3 is not about letting strong players through, it's more preventing weaker players from getting lucky. The results of the experiment do not reflect that. In this type of experiment 2/3 doesn't matter, because each outcome is already decided. 2/3 is supposed to prevent the quick one turn wins and the random bad hands that players tend to get regardless of skill level.

 

[NoPoke]

erm the premise for introducing 2/3 is that it reduces the luck in the tournament . So I took it to the extreeme and removed ALL luck except the pairings. This shows that introducing 2/3 will not prevent players complaining about bad luck.

Your argument for introducing 2/3: that it allows you to recover when a weaker player gets lucky is flawed. However if I accept your statement as true then playing 2/3 will determine who is best in nearly all matches. Which brings us right back to the idealised experiment where the best player wins each match. I could roll my hundred sided die a couple of times each round and have the weaker player get lucky.. but this serves no purpose since such action will just increae the likelyhood of stronger players having a worse match record.

Unless you go to absolute extremes weaker players WILL still get lucky. Even after 10 rounds with the best player almost always winning say best 16/30 the pairings luck still persists and will still cause accusations that player xyz got screwed.

I completely fail to see how increasing the luck in the tournament from the idealised simulation does anything other than increase the chance of a weaker player getting lucky.

The cut is set to allow the majority of strong players to have some bad hands and some bad luck and still make it through. The number of swiss rounds is higher than the minimum number to just get 32 players with a 3-0 record. Those extra rounds allow players to make the cut even though they have lost a couple of games.

I supose I shouldn't be surprised that players won't let go of fondly held beliefs even when the evidence suggests that those beliefs are flawed. The T32 is the important tournament. The swiss is just to reduce the field and select a good majority of the stronger players. 7 rounds of single game matches will achieve that. yes some players will get lucky, yes some very good players wont make it. Adding 2/3 into the swiss wont change that. I know that very few of the posters here are listening but fwiw I view the T16 as the REAL tournament the first round of the cut is still eliminating the weaker players who got lucky. So if you want to spend more time on the tournament save it for where it is needed and leave the swiss as single game matches.

[ There is merit in increasing the number of swiss rounds from 5 to 7 but mostly because it allows the best of the best players an inproved chance of making the cut. These best of the best would be perhaps less than half of the players in the cut after round 5. The remaining 16 places would be a mix of good and wild cards. There is no merit in using the available time for the swiss rounds as a smaller number of 2/3 rounds.]

 

[Articjedi]

no, but even if your results are true in this simulation, it would be less of those t32 players making elimination if there were single games instead of matches.

The results in this type of simulation would be no different if you did single matches, since you already assume the best player wins. You have to add some type of luck factor in, probably some type of disadvantage when determining winners of matches, then try 2/3 vs. single elim and compare who gets in, there should be a difference in that case.

 

[Flaming_Spinach]

Arctic Jedi is right, why don't you give the higher-ranked player an 80% of winning or so, then see how it all stacks up, in both single and match play.

 

[GrandmaJoner]

Yeah NoPoke this fake thing you came up with is completely invalid, and your arguments still havent changed from the very first time I posted this idea. But if you've been following the posts, you'll notice that I've also proposed to increase the amount of rounds, or to extend the round length by 15 minutes. Those are probably more likely to happen this year.

 

[PokePop]

Whoosh!

You guys are not getting what he is saying at all. He's not trying to prove that single games are better than 2/3.
What his simulation proves (not what he supposes or guesses, but proves) is that swiss rounds will produce a result where some randoms get in and some top players don't just based on the very structure of how swiss is run. It doesn't matter how you run each match in the swiss round. They are still all single events in the swiss event.

The only way to get "better" results would be to try to come up with some other way of determining the top 32. Just playing more games in the same tourney structure is not going to do it. The math does not lie.

 

[Flaming_Spinach]

That is very true Pokepop, but Nopoke has neglected the fact that a person ranked 1st could lose to the person ranked 128th in the first round (probably due to bad luck). Nopoke has neglected the fact that some sense of luck even exists.

What I'd like to see is the same program run again, but this time give the higher ranked person an 80% chance of winning, instead of a 100% chance, this way we can at least account for some sense of luck.

If the results turn out to be the same (or close) even after this, than we can be assured that luck does not play a very large role in swiss-style rounds.

 

[Articjedi]

The way I see Nopoke's program works like this:

In 2/3 the better player will always win the match. But even if he did this with single games the better player will always win the match. Of course the results wouldn't be different, because the outcomes are already determined. That's why his model is flawed.

A better model is this. It's like if you flip a coin. let's say the coin is rigged so you get 75 percent heads, to reflect a good player playing a less than good player. if you only flip once, ie playing one game, you get a 75 percent chance of victory for the better player. However in 2/3 that changes to 84 percent. So in worlds that means every round the better player will win 9% more games, about 12 games if there's 128 in that division. Each of those games gets added to the better players score, so now better players get into elimination

 

[The Gorn]

Correct me if I'm wrong NoPoke, but the better player ALWAYS wins in your model, right?

So what you've done is shown how the Top 32 is affected by pairings via TMS, resulting in
8 (or 6) of the Top 32 BEST players not making it.

Interesting.

Maybe the first order of business should be evaluating TMS and how it handles pairings.

 

[PokePop]

That is very true Pokepop, but Nopoke has neglected the fact that a person ranked 1st could lose to the person ranked 128th in the first round (probably due to bad luck). Nopoke has neglected the fact that some sense of luck even exists.

What I'd like to see is the same program run again, but this time give the higher ranked person an 80% chance of winning, instead of a 100% chance, this way we can at least account for some sense of luck.

If the results turn out to be the same (or close) even after this, than we can be assured that luck does not play a very large role in swiss-style rounds.

That's not the point of what he's trying to do.
He hasn't "neglected" it.
He purposly took out all the luck of everything other than first round pairings so examine how the results would look with the least amount of luck possible.
Any luck introduced will just produce more "randoms" getting into the top and more top players "getting gypped". He doesn't need to do what is a heck of a lot of work to show that it can only go downhill from the optimal results he generated. Once you start introducing "unfair" losses, you quickly get to the kind of results that you all are trying to avoid. So what would be the purpose of running that simulation?

 

[PokePop]

Correct me if I'm wrong NoPoke, but the better player ALWAYS wins in your model, right?

So what you've done is shown how the Top 32 is affected by pairings via TMS, resulting in
8 (or 6) of the Top 32 BEST players not making it.

Interesting.

Maybe the first order of business should be evaluating TMS and how it handles pairings.

Exactly.
A much better line of inquiry, rather than playing around with percentages of luck.
What other formats are there for running the rounds of a tournament?

 

[NoPoke]

Yes the better player always wins in the simulation. Player A1 is always going to beat any other competitior because that player is the best. Sadly player E128 being the worst looses to everyone.

What I've shown is that

The random nature of pairings in swiss, especially in the early rounds, results in some strong players not making the cut.

Increasing the number of rounds helps but does not eliminate the problem.

There is a degree of instability in who makes the cut. Each extra round swaps some of the players who were only just in with some who were only just out.

I have shown that pairings luck is highly significant.

-------------

I have not shown that 2/3 is bad.
I have not shown that single game matches are bad.
I have not indicated any way around stalling.

I have a great deal of sympathy with Joners original post in that he identifies two areas that we should always be mindfull of. Stalling and luck. Both issues are addressed in his proposal by increasing the ammount of time allocated to the swiss tournament. I do not believe that this approach is the most efficient use of time, because both problems still persist even if double the time is allocated to the swiss. However I do aplaud the effort that he has put into the original post.

Problem A How to fix stalling?

Problem B How to reduce the impact of luck?

Problem C Time is a finite resource, how to spend it wisely?

Joner's proposal is threefold:
Joner proposes increasing the round time from 30mins plus overhead to 60 mins plus overhead
Joner proposes increasing the number of rounds played.
Joner proposes introducing 2/3 play


the difficulty with the proposal is that Stalling is only partly addressed. Luck persists, and the organisers time constraints have been completely ignored. It is thus too easy for POP to dismiss the proposal. I am not saying that the proposal is without merit just that I don't see a strong enough case for the changes suggested.

If anything the changes that POP has made to OP make stalling a bigger problem that it once was. No more draws and a timed win counting the same as a full win both encourage stalling. In single game matches the stalling problem shows up mostly in the last 5 minutes of a match, this is not the case in 2/3 play. I can see that this is why Joner made such a BIG increase in the round timings along with introducing 2/3 play. doubling the time for each round is how Joner is trying to avoid making stalling worse when 2/3 is introduced.

POPs current penalty guidelines that slow play gets you a caution don't help either.

The tournament structure being used by POP is a hybrid. A number of swiss rounds followed by a cut. The swiss establishes the seeding for the single elimination tournament that follows. That the swiss tournament is flawed is accepted by (almost?) everyone. It is the precise nature of those flaws and the impact that they ACTUALLY have on the seeding going into the single elim that needs to be examined if we are to present POP with a powerfull argument as to why the swiss should change from 7 rounds of single game matches.

In Chess the first round pairings are seeded based upon ELO rating.
In Chess who goes first is not determined by luck alone.
We can learn a lot from chess. But that doesn't mean we should slavishly follow how Chess resolves its problems. This is Pokemon and not Chess or even Magic the Gathering.

 

[NoPoke]

I agree that my model is flawed but probably not for the reasons that some have suggested. I had no intention of modelling the real world situation since I cannot place sensible figures on the various lucky components. So I eliminated tham and instead anaylsed just the impact of the random early round pairings and how increasing the number of rounds reduces the impact of those early round pairings.

So my model is flawed because it is idealised. But it is not seriously flawed in that it does expose the effect of pairings and number of rounds which was what I wished to highlight. 10 rounds and still some of the best players dont make the cut......

[under Joners proposal 4 rounds of matchplay take the same time as 7 rounds of swiss]

 

[bullados]

What about bringing back a "points" system in which you'd get 3 points for a straight win or win through concession, 2 points for a prize win, 1 point for an overtime win, and 0 points for a loss? This would help eliminate the stalling issue because, by stalling, you're actually screwing yourself out of more points and possibly a spot in the top cut. Realistically, this proposal does nothing to reduce the impact of luck, but I also believe that luck is an important part of the game, and MUST be realized in order for the game to be considered competitive. Time constraints might become a little bit more stretched, mostly due to the amount of time taken in between rounds for recording and pairing of the new rounds, but, overall, it might add about 15 minutes to the total tournament time, especially if the TO has enough IMPARTIAL staff in order to properly guage the different types of winning and scoring those different ways appropriately.

Please don't flame me, it's just an innocent idea...

 

[NoPoke]

Bullados I've been trying to figure out a revised points system too. That early lucky FTKO is the stumbling block. Will players complain noisily like they did about draws when a lucky FTKO gets you more points than a 5 prize lead when time is called. Unfortunately I think I already know the answer.

I was trying to figure out a points system that would allow for the FTKO but would not increase round timings. In essence tring to find a way of introducing 2/3 to 30 minute games! Not that it is true match play rather that players get a chance to earn some points when they suffer a loss when the match has hardly begun. The points system has to cope with many many matches being unfinished at the end of 30mins. I didn't like any of the ideas I came up with. There seems to be no option but to increase the time allocated per round significantly if match play is introduced. (unless someone out there knows better )


Maybe reducing the points benefit for winning by benching your opponent when you have yet to take a prize. ie penalise the FTKO.

3 points for a full win (ie. all prizes taken)
2 points for a prize win
1 point for winning by benching your opponent when neither player has taken a prize.

 

[PokePop]

NoPoke: What about my question: What happens to the standings if every round of the swiss is random? How does that impact some top players making the cut?
My hypothesis is that it will result in fewer deserving players being cut because those near the lower limit will not have been as likely to have had to play the top players, but will have had more games against scrubs, therefore giving them a truer record.

 

[NoPoke]

PokePop, Since pairings are completely random rather than the swiss approach of always trying to match players with similar strength you end up with slower convergence on the best 32 players. I believe I'm correct in saying that 1 in 7 tournaments you still have more than 32 players with a perfect record even after 7 rounds of play. [I was hoping that someone else would be able to confirm or refute this statement without having to think too hard about it myself. Just being a little lazy ]

A straight swiss tournament is designed to establish a single winner in as few rounds as possible. The problem with random pairings in the early rounds is overcome by introducing rating based pairs for the early rounds. This is not unlike the approach in the STSs where some entrants received 2 byes. But those were open tournaments I believe not that I was ever able to attend one

using random pairings in all rounds removes the instability that is inherent in swiss pairs. The instability I'm refering to is that as each round is added to swiss some players just above the cut are exchanged with some players just below the cut. random pairings will I believe converge more slowly but without the instability.

So if you are going for a T32 cut then straight swiss even with the random early round pairings will establish the top 16 players rapidly. The remaining players 17 through 32 will be mostly correct but will oscillate about the true result.

[for anyone with a maths bent this is not unlike the effect that over using 'accelerating factors' have on the convergence of iterative methods of solving the roots of equations. ( Newton-Raphsons method springs to mind ) ].