thoughts on tournament organisation
Posted on Dec. 29, 2018, 7:36 p.m. by Winterblast
Multiplayer Tournament Organisation - theoretical background and practical implementation
When the interest in EDH became high enough in our local scene, the idea of testing our skills in a tournament came up. Obviously the first step was to check if the LGS was interested and the first answer when being confronted with multiplayer EDH was something like “there's no option for multiplayer in the program that is used to organize tournaments…how would a multiplayer tournament plan even work?”. That first reaction got me thinking what is actually needed for multiplayer tournaments to be hosted easily, without a lot of work and possible problems in organisation and execution. The first reaction of the LGS made it clear that one important aspect obviously is to have a program or at least an easy-to-use tournament plan to calculate pairings and points to get a final ranking. I had a look and tried to find multiplayer tournament plans on the internet, but nothing was even remotely usable like a 1vs1 swiss system. Also, trying to adapt 1vs1 swiss system for multiplayer seems like a possible approach but it comes along with several problems you definitely want to avoid. More about that later.
Before we can think about how to make a user friendly interface for a tournament system we need to determine what we actually expect from a multiplayer tournament or a tournament in general, which problems could arise in a multiplayer setting and how to prevent them. For a multiplayer tournament to be fair and interesting I believe some basic expectations are:
1. It has to end within a reasonable time
2. It has to compare the playing skills of the participants in a correct way that produces a clear ranking
3. The initial pod constellations should have little to no influence on the final result
4. Intentional teamplay and unintentional kingmaking has to be prevented by the system
5. The tournament system should motivate players to stay in the event and play for the win until the last round
A tournament has to end within a reasonable time
This is a pretty obvious requirement for any tournament - if you start sometimes in the afternoon or evening you want to be done with it on the same day. It's not hard to come up with a tournament system for multiplayer that provides a clear ranking in the end, but to compare a reasonable number of players in pods of 4 many ideas I have seen so far either don't fulfill other expectations we should have towards a tournament, or it takes far too long to finish.
If we consider how long even competitive EDH games can take sometimes, planning an event with more than 4 rounds can become a problem in real life. A time limit per round is also reasonable but it would have to be higher than for 1vs1 matches to avoid getting draws frequently. The tournament system then also needs to be able to calculate logically correct pairings for each round and produce clear results in the end, even if one or more rounds end in a draw. Leaving the round time open is an option but having 4 rounds can already be a practical problem when just one match in every second round takes incredibly long.
Let's assume 1.5 hours maximum per round with 4 rounds in the event, that's a maximum of 6 hours without breaks between rounds. That's still reasonable for an event at an LGS or similarly small venue. Within these rounds the participants have to be compared in a way that is fair and related to actual performance, not how people were randomly seated.
A tournament has to compare the playing skills of the participants in a correct way that produces a clear ranking
This sounds easy but if you check upon what people have come up with for multiplayer tournaments, there’s a lot of irrational and downright stupid concepts floating around. The first important point is to only reward winning a whole round. The system should never motivate players to aim at something other than winning, not kicking single opponents out or avoiding being the first one to die - the goal is to win against all opponents and points shouldn’t be given for anything else. But how do you get a clear and replicable ranking after 3 or 4 rounds (depending on the number of participants)? Anything that takes longer will be impractical or even impossible because of time constraints. This is the first aspect in which various adaptations of 1vs1 swiss mode fail in a multiplayer setting.
Instinctively, because we know that from 1vs1 tournaments, we would like to put the winners of round 1 together in round 2. If you follow that procedure until the final round, you will see that losing will pair you with other losers and winning against losers is worth less during the course of the tournament - not because you get fewer points for these wins but because you are already in the losers bracket and won't be paired against a previously better performing player later anymore. On the top you will have one winner and one loser in the final round and these are automatically 1st and 2nd place. If you used the same principle for multiplayer you would have a final round that produces one winner and up to three losers that might share 2nd place.
Pairing winners together for round 2 is also awkward when there is an odd table number, for example with 12 players (3 x 4). Round 2, if you want to pair winners together, would be 3 winners and one loser at a table and then it would have to be a random decision because there's no logical explanation for which loser should take a seat at that winner's table. Even with an even table number that kind of random seating needs to be applied at some later point in the tournament if we adapted Swiss system because there will be too many players with an equal number of wins/points. We do want to reduce randomness and provide a fair setting though.
In order to fix that, a reasonable solution is to place every winner of round 1 at a different table for round 2. To keep the logically correct aspect of the Swiss mode, that winning against losers of previous rounds is worth less than winning against winners, we can rate how difficult it is for every player to win at the table they are currently playing against. This has to be calculated from how the opponents have performed in the previous rounds. That points us directly towards ELO rating systems, which do exactly that: reward points according to how hard a match was to win. However, we can't just take the exact same system as it is used for example for Chess because the calculations are made for 1vs1 games there.
The initial pod constellations should have little to no influence on the final result
This aspect is very much relevant for multiplayer tournaments. Being randomly placed at a table with a possibly triple bad matchup is much worse than it could be in 1vs1 matches. If an elimination system was used, the random initial seating could already end the tournament for unlucky players.
What the tournament plan should measure in multiplayer is deck consistency against a large variety of possible matchups in the meta. Considering the nature of the format, it makes sense to let all participants play against various table constellations before the chances of getting a higher ranking are completely lost. If you take into account the time restrictions and the need for a clear and replicable ranking, which have been discussed above, there aren't too many options left for how to arrange pods in the rounds after the first.
A method that measures deck consistency throughout the whole event and reduces randomness linked to the initial seating would be this: after the first pods have been randomly created, the following rounds alternate between balanced and stratified pods. In rounds with balanced pods, the currently best players are distributed equally among all tables, which makes the total number of points among opponents (almost) equal at each table. Stratified pods are arranged by ranking, meaning the current ranks 1-4 form a pod, then 5-8, 9-12, and so on.
This means that if you win early you will have to defend your position in later rounds against tables that also contain opponents that haven't played as well in the previous rounds. You basically have to prove in every 2nd round that your early performance wasn't just because the random matchup was favorable but that you can really win consistently against the whole field. From the other perspective, if you lose early you still get to compete against the better players of other pods later in the tournament. For example a previously winning storm player might run into a table of stax decks that have been beaten by some midrange stuff in the early rounds…in a Swiss or elimination system their losses might have led to them not crossing the storm player's path even once in the whole event, making the randomly assembled round 1 have a big impact on the final results.
Intentional teamplay and unintentional kingmaking has to be prevented by the system
One of the main reasons for why the idea of multiplayer tournaments seems to be problematic for a lot of people is that they believe that collusion can’t be prevented and that no judge could correctly rule when something was intentional illegal teamplay or just a misplay. To target collusion with how the tournament is organised, we need to have a look at how people would realistically try to get an advantage from illegal teamplay in a free for all tournament.
To actively help another player in a multiplayer event you would mostly have to sit at the same table and the assistance you can give is limited to that round. One method of helping another player win is for example to play worse than you could or to focus on the other opponents instead of making strategically correct decisions. Besides letting a judge decide if a move was intentionally made to help another player or just a misplay, we can try to discourage the behavior by using a tournament system that makes it very hard or even impossible for the helper to know what the outcome of the teamplay will be in the end. If we use an elimination system, helping another player win even one game would have a huge impact on the further progress of that player. By using the method described above (rounds alternating between balanced and stratified pods) it's almost impossible to predict how the helper would have to behave in the other games to further help his partner get a better ranking and enabling only one win wouldn't really do much for the overall performance in the tournament either. If someone still tries to do that it won't have a big impact on the tournament but it's actually unreasonable to try and collude in that system in the first place.
What is probably the most realistic scenario is when kingmaking happens through legal game mechanics. I'm not talking about someone taking a risk that didn't pay off or someone misjudging a situation and making a play that turns out to have helped an opponent…that's the nature of multiplayer and table talk is important to prevent especially less skilled players from handing out free wins frequently. And you are sure to have these people with poor threat assessment or little understanding of the format enter a public competition - it's not a problem though, multiplayer needs in-game communication even in a tournament (not from bystanders obviously).
So, what is the kingmaking problem that needs to be addressed by the tournament system? It's people leaving the game before the round ends. Conceding is possible at any time according to the rules, even when spells/abilities are on the stack, and it makes perfect sense in 1vs1 matches, especially when you consider tournament play. In a match that requires 2 wins and has a time limit it makes indeed sense to forfeit one game that seems to be unwinnable and would take really long to finish. You will need the remaining time to play the other games of the match and maybe secure a 2:1 eventually. However, in multiplayer it's not “I lose, you win” when someone concedes but it's one of 3 remaining players and it's more or less predictable who of them is favoured when a certain player leaves at a given time. And as we play for one win per round (time reasons, remember?) it's even worse when even a single game turns into a lottery all of a sudden. Stax pieces suddenly disappearing, a lower number of opponents, controlling effects ending at instant speed, spells fizzling because of a vanishing opponent…
In a tournament it is assumed that players do want to win and avoid losing and base their actions on that goal. As we don't have rounds with more games, losing on your own behalf in that setting is a completely irrational move and that makes the remainder of the game random, which is what we don't want in a tournament. As you don't help yourself by conceding in multiplayer, it can be assumed that you actually want to help another player with winning over the other 2 opponents and that would be collusion. Therefore it's in the best interest of everyone to not allow conceding of individual players in a multiplayer tournament. Obviously, if someone has to leave the whole venue in a hurry for external reasons, they would drop from the whole event and the possibility of collusion can be excluded.
The tournament system should motivate players to stay in the event and play for the win until the last round
This point is linked to the question of how to arrange pods in the various rounds of a tournament. Players leaving the event will cause the possible pod constellations to change. I've had such a case in practice and the result was that instead of 3 tables with 3 players there were only 2 tables but with 4 players each after the 2nd round. This doesn't ruin the whole event but it requires some random seatings again in the following round - more details later in the practical part.
A tournament system can't prevent people from leaving the event because of external reasons. However, if we assume that players in a tournament want to win, it is possible that someone drops from the event as soon as they realize that they don't have any chance of reaching a rank with prize payoff. In an elimination system this is automatically done by the system, in a Swiss system players will also see if they have no chance of winning anything at a certain point. If they don't drop from the event then, they might be encouraged to make random or downright kingmaking moves because they know winning isn't possible anymore and they might think it's funny to tilt games instead. Having a tournament plan that allows for people to reach ranks with prizes until the later rounds is an incentive for staying in the event and playing as well as they can.
The following part will address how to implement the presented points in a working tournament plan and how to use it in practice.
Creating a multiplayer tournament plan for practical use
Taking all the points above into account, a tournament system has to be created that removes the possible problems of a multiplayer setting or reduces them to a minimum. Maybe you will remember December 2017, when the rule committee made silver bordered cards temporarily legal to play. Maybe you can also remember that this sparked a discussion about seperating competitive play from the rule committee. When it was clear that the RC didn't put out more irrational and whacky changes to the rules and the banlist was updated twice with no changes, the discussions around an alternative rule committee basically stopped and turned towards multiplayer tournament organisation instead. These discussions provided a concept for a tournament plan that I have tried to put into a usable interface with some adaptations. I am using it in practice for our local tournaments, which we now host privately.
The tournament plan works for 6-16 players. A tournament will take 3-4 rounds to complete and produce a clear ranking. Adding more players is possible but it would take more rounds to get a clear ranking when more tables are needed. This may be too long to execute, so it makes sense to run seperate groups and combine them in the final round. Even bigger tournaments could easily be managed that way, so I will focus on how to run the tournament with 6-16 people.
First of all we have to determine how many tables we need because that will tell us how many rounds need to be played. 4 player pods are preferred but for many numbers of participants we will also need 3 player pods. For example with 6 players we have only 2 pods of 3, with 12 players we have 3 pods of 4 and with 14 players we have 2 pods of 4 and 2 pods of 3. For 2 tables we need only 3 rounds to get a ranking, for 3 and 4 tables we need 4 rounds. It has to be noted that players dropping from the event might change the number of pods that can be formed but the number of needed rounds that has been determined by the initial pod constellations will not change.
As stated above, we will look at the ELO rating system for inspiration. However, if you have a look at how to calculate the points there, you might have understood how it works in Chess for 2 players but putting 2 more players into the equation is a horror to calculate and highly impractical when it needs to be implemented for example in excel. The ELO system rewards points according to how hard a game was to win - if you were already performing better than your opponent you get fewer points because it was more realistic that you win against that opponent. The opponent loses that amount of points then. In the system we use, we only calculate the gain for a win but nothing will be subtracted for losing. The calculation of the gain will also be done differently because we need to calculate the difficulty of the whole table (=all 3 opponents), not just one opponent.
Players are seated randomly for round 1 and everyone starts with 10 points. A win will reward half of the average opponens’ points, which means in round 1 it's always +5. Round 2 are balanced pods, which means the points on each table are equal. In that round we try to maximise the number of opponents everyone plays against, so the winner stays at the table and the losers in turn order move 1, 2 and 3 tables ahead (after table 4 comes 1 again in that cycle). With fewer tables, when it's not possible to create completely new matchups, the player diagonally across the winner (the player to the left in 3 player pods) stays at the table as well and only the other losers rotate.
In round 2 we have tables with one winner and 3 losers, so the points are 15-10-10-10. The possible gain is now different for everyone as you will get roughly 5.8 points for winning against the winner of round 1 and the two losers but the winner will only be able to add 5 points to his balance for winning against 3 losers.
If we stuck to the aspect of the ELO system that the total number of points stays the same, we would have to divide the gain in a way that is relative to what every opponent could have gained in order to subtract the correct amount from the losers and that is a huge mess to calculate. Therefore we only reward the points and leave the losers’ points as they were. To reward deck consistency we do reward one additional point for each consecutive win and consecutive losses are punished by subtracting one point. This will also result in incremental differences between players with seemingly equal points in the middle field and acts as a sort of tie breaker for the final ranking.
Pods for round 3 are stratified and assembled by putting the 4 currently highest ranking players at table 1, the next best 4 at table 2 and so on. The pods in round 4, which is the final round, are balanced again. This round is assembled by sorting the players by points and then taking players subsequently from the top and bottom of the ranking and placing them together until the tables are full. If you assign table numbers for a ranked list of 16 players, they would be like this then: 1234123443214321
If you run a tournament with multiple groups, you would now play another round with stratified pods but take the top players from all groups to fill the tables. Players of each group ranking lower than 4 would not have to play out that 5th round because at that point there's no way they could get into the top 8 anymore and even with 32 players (2 groups of 16) there probably wouldn't be prizes available for lower ranks.
I have implemented this tournament plan in Excel as I wanted to have all the point and pairing calculations done automatically, but you could just as well use pen and paper if you don't have a laptop with you where you host the tournament. I wouldn't recommend manual calculation because it's really a lot to calculate and mistakes might happen. In the Excel file I use, all that has to be done is to enter the results of each round (different number codes for win, loss, consecutive win/loss) and the pods for the next round can easily be assembled in another sheet by sorting the players by points, assigning table numbers according to ranking (stratified pods) or as explained above for balanced pods. Then sort by table number and the players appear at the correct tables in the 1st sheet. This should be user-friendly enough to be used at an LGS or for similarly sized private events.
The next step for me is to bring the possibility of draws into the tournament plan (and with that the option of imposing a round time) by giving a draw a number code to be entered as a result. This should award half of the “gain” to each player, which is in line with how points are given for draws in the ELO rating system. The reasoning behind that is that not only a win but also a draw is harder to achieve for the currently lower ranked player. The higher ranked players will get fewer points for a draw, so they have no incentive to go to time (intentional stalling is a job for a judge anyway) by playing overly cautiously.
If anyone wants to use the tournament plan in Excel or needs some help with how the plan has to be used when there are also tables with 3 people, feel free to contact me. You can find me on Facebook as Walther Winterblast, on tappedout.net as Winterblast and I'm also available on discord as Winterblast#0881.
This is a huge writeup and I want to remember to back to this. Will update later when I've collected my thoughts and reread.
I'm stuck somewhere between cautious optimism and resignation in the face of futility when it comes to serious play for EDH in an organized setting; introducing payouts for winning complicates things for multiplayer EDH to a tremendous degree.
December 29, 2018 9:47 p.m.
Wow. I would like to implement this in an event that I am currently planning and then return with the results.
December 29, 2018 11:34 p.m.
Arvail the proposed tournament plan is what I use for our local events. We host them privately now to be more flexible with trying out new stuff than we would if an LGS hosted them. It has been fine with relatively big prizes because we only pay out something for the first 3 places. It would be cool if more people tried to use that plan and see how it goes.
December 30, 2018 6:34 a.m.
I've now read this on multiple occasions. Basically I don't think that it's possible to run a decent tournament and/or league for EDH. Ignoring the feasibility of problems 1-3, there's just no way of ever properly addressing 4 and 5.
When it comes to problem 4, I think you recognize how insane it would be to try to parse through lines of play in some grand effort to sniff out collusion. Doing something like that just isn't going to yield good results. It's also going to be a major time sink. As a result, we're left to polish an exceptionally small area in which problem 4 presents itself.
I also think it's a bit silly that your main incentive in addressing problem 5 (payouts possible for players even later on) directly contributes to problem 4.
Still, I could sit here and theorycraft the feasibility of this rule set to death, but I think it's major problem is that it's simply too complex. Actually implementing anything like this just isn't realistic for stores and getting players to understand the nuances of the system would just take too long. Outside of hyper committed groups, this can't take off.
January 4, 2019 9:41 p.m.
Arvail to address collusion I think we have to understand what makes people do it in the first place. The underlying idea is basically to team up with another player because it seems like the event can't be won alone and that there's a realistic chance that one player could help the other win...to share the prize in the end. If it's not realistic that you really help another player a lot by losing intentionally in one round, the motivation for giving up your own win drops significantly.
What the system does, is that it takes away chances that one player might give another a bigger boost by illegal teamplay (because of the rating and pairing system instead of elimination/swiss) while giving more incentives to play for your own win even later in the event (because you aren't automatically out of prize range if you lose one round). It's not like people wouldn't still be able to try, but they won't get the results they would like to have when trying to work together.
For practical use, the players don't need to know ANYTHING at all, they just play. For the host, the plan is rather easy to use, even in this early stage, made with my own excel knowledge. All people have to do is to follow the instructions for pairings given by the host and report back results after each round. No one needs to understand any of the background calculations and I wouldn't try to explain how exactly the concept of elo rating has been adapted because it's probably too difficult for the average player. Not even the host would really have to understand what the excel sheet calculates in the background if he just knows how to use the sheet. This was one of the main reasons for coming up with something that people can just take and use like the many options you can find online for 1vs1 tournaments.
January 5, 2019 12:10 p.m.
You're still stuck on the whole notion of scooping for assistance. From my experience, I think it's far more likely that we get into issues with target selection. People constantly argue that politics goes down as threat assessment will dictate decision making more and more at higher levels of play. While that's true to a certain extent, that is based on the notion that players have a high level of skill (not a certainty) and that they won't make risky plays for bigger payoffs later on.
For example, I might have a way to push 5 or so damage consistently to one player without opening myself up in any significant manner and could end up killing them across 3 or so turns provided no drastic changes happen. My opponent may be in a decent position and could potentially combo off. I could take that player out, but then my other opponent, who has a dominant position, would be quite free to finish me off. There is no right answer to this situation. The decision to wait on killing a player may balance the board a tad more over the next coming turns, but it could also end up granting that player a window to close the game.
This is a regular occurrence at even high level tables. Once you add in prize support to incentivize collusion, things get incredibly murky. To add to this, it's pretty common for people (even at high levels) to target based on personal reasons. If there's really no way to determine which of two opponents is a greater threat, people default to attacking people they don't like.
I could go on, but just barring scooping for king-making purposes fails to account for collusion in all the various places it crops up in any meaningful fashion. The idea behind collusion doesn't even have to be to notably increase your chances of winning because engage in the practice has such a low opportunity cost that not doing so is just suboptimal. It's going to happen.
I would argue that proper understanding of the ruleset you're expected to abide by is instrumental in creating an experience that truly measures player ability and deck construction. If people aren't sure how the tournament even works, its credibility is called into question. Worst yet, it might actually incentivize collusion even in cases where the benefits of the practice don't actually tilt the tournament in the favor of the ones colluding. Not understanding the ruleset will also make the authoritative position of organizer shaky.
January 5, 2019 1:16 p.m.
Arvail what I said in the last comment wasn't about the conceding aspect. That's just one point that mostly prevents games from unintentionally becoming random by people leaving and sometimes this could also help a certain player and might be part of a illegal deal. However, that's not what you are talking about here.
Helping a player winning a round means that you have to be in that same game. It also means that you voluntarily give up or reduce your chances of winning a prize because you think that with your help the other player will have such a high chance of winning that you two can share the prize eventually.
If you look at how the rating system works, you might be paired with your partner in crime in round 1 and help him win. In round 2 you won't be able to help him because you will sit at another table and in round 3 you also won't see that player because if he won round 2 as well he will sit at another table in the ranked pod round. If he loses, then your chances of getting a prize that is worth sharing are already low. You might see the player in round 4 but it's almost impossible for you to predict which rounds you would have to win or lose in order to see him in the final round. This is what discourages teamplay most imo because it targets the motivation for collusion. In order to win a prize, the player being helped would still have to win the biggest part of the event on his own and why would that player need to team up in the first place? The other perspective is why would the helping hand give up their own chances at winning anything when the correlation of teamplay in one round and the outcome of the whole event is so low?
The point here is that people might try what you described, but it won't be much help, unless the better player wins all the other rounds on his own anyway. If it's not planned and people just make bad moves because they don't know any better, that's not collusion then. If someone would unknowingly win the game for someone else, that's upon the opponents to manage - in multiplayer games in-game talk is necessary and you also have to account for other people's misplays or seemingly unlogical moves. That's part of the playing skill in multiplayer to decide whether an opponent is just playing badly or has some trick up his sleeve that you don't know and talking is one way to find out.
As for understanding of the tournament system, it's not affecting the games you play. The rules are clear because every round is a FFA multiplayer game and people know the rules of the game. The host can explain that throughout the event you collect points for winning against other players and that the amount of points you gain depends on the current standing of the people you play against. I'd argue that it's not relevant to actually know that it's half of the average of the opponents at the table, but it's an information you could give briefly. How pairings are calculated is definitely irrelevant for the players because if you play 1vs1 at a shop the pairings for each round are also just given on a display or announced verbally and you have to take the place you are assigned.
January 5, 2019 4:46 p.m.
Why are you assuming that collusion is limited to two players? I've seen a regular 4 man group of friends who normally play by themselves enter a 16 ish man tournament before. At that point, the priority in rounds often becomes eliminating the people doing the best outside of the colluding group.
January 5, 2019 5:03 p.m.
if it's more players they will realistically end up being distributed on different tables because of the random initial seating and the following rounds will also mix them up in a way that is unpredictable for anyone of them. It's basically impossible to secure a win for one specific player in that way. You would realistically have two of them meet here and there but eventually, if they need to split the prize among 4 people, there's the question if it's worth doing that in a system that makes it hard to predict the outcome for the group of colluding players. there will be rounds in which they don't meet an they have to play these as well...and it's definitely not predictable if losing or winning these rounds would help or hurt one of the friends in the long run.