Home | About | Tournament Winning Decklists | Forums

Swiss Rankings


#1

Is Total Prestige the most fair way to rank players after N rounds of Swiss?
Would a SoS Weighted Prestige make a little more sense as a ranking metric for Swiss tournaments, to take into account the quality of the wins attained?

Under current Swiss scoring systems, players in a tournament are ranked by:

  1. Total Prestige (2 points for a win/bye, 1 for a draw/timed win, 0 for a loss/timed loss)
  2. Tiebreakers on SoS = Sum(opponents prestige)/(#rounds^2)
  3. Additional tiebreakers on xSoS = sum(opponents opponents prestige)/#rounds^2

Pairings are determined in a pyramid way whereby players are ranked with those around them in terms of current prestige after each round i.e. winners are matched with winners, losers with losers. This naturally causes a situation whereby those who do well in the early rounds, will have relatively harder matches in the later rounds as they are paired off against eachother, and consequently building a stronger SoS than the rest of the field, but their prestige becomes increasingly harder to come by as they progress up the pyramid. In some ways the swiss systems punishes you for success whilst simulataneously rewarding you for failure in making your next round harder/easier respectively. This kind of catch up logic is apparent in the pairing system, but not in the final ranking system itself (except in the case of tiebreakers).

Often some of these players who have been playing in and around the top tables for the duration of swiss will miss the cut, replaced by someone who has relatively had a much easier ride on the mid tables who effectively sneaks in on the final swiss round. Whilst the strength of players faced is taken into account only in the case of tiebreakers, the Total Prestige method does not effectively weight the quality of the wins adequately. In other words all wins are equal under the current system.

One proposal could be to devise a metric weighted by SoS such as (Total Prestige* Sum(opponents prestige)/(N^2))/N = SoS* Prestige/N, using Total Prestige as the tiebreaker thereafter.

This has the property whereby the Prestige earned by a player is weighted by the average quality of the opponents attained he/she has played against over all rounds of swiss. For example, if Player A had scored 12 points and Player B 14 points in a 5 round tournament, and their SoS was 2.4 & 2 respectively, then Player A would finish above Player B on this metric (SoS* Prestige/N) with Player A obtaining a score of 5.76 vs Player Bs 5.6.

Running the numbers on the Worlds 2015 Swiss leads to some interesting changes. Lucas Li, whose phenomenal performance to accumulate 22 prestige in the context of a 3.05 SoS is recognised as the 3rd seed in Swiss, whilst Jen Erickson & Chris Dyer would also sneak into the cut on based on their SoSs of ~2.65. Notably a few of the late climbers such as Dave Hoyland would not make the revised cut, which could also be seen as a criticism that using a SoS weighted metric leads to a top table “club” which has its own intertia making it hard to break into if you have lost a few early rounds, increasing inevitability and reducing some of the excitement of the tournament. Conversely to this, one could argue that players who have been in and around the top tables all day, are to some extent are more deserving to be in the cut at the end, and a SoS weighted metric makes that more likely.

Interested to hear peoples thoughts.

Worlds 2015 Swiss (top 24 based on new rank are shown):

original rank name prestige sos prestige*SoS/N prestige new rank rank difference
1 Brandon Hauk 30 2.69 10.09 30 1 0
2 Minh Tran 26 2.61 8.48 26 2 0
16 Lucas Muxi Li 22 3.05 8.39 22 3 +13
5 Tim Fowler 24 2.69 8.07 24 4 +1
3 Dien Tran 26 2.42 7.87 26 5 -2
6 Noah McKee 24 2.61 7.83 24 6 0
7 Stjepan Pavuna 24 2.46 7.38 24 7 0
8 Gary Bowerbank 24 2.45 7.35 24 8 0
9 Dan D’argenio 24 2.39 7.17 24 9 0
10 Timmy Wong 24 2.39 7.17 24 10 0
4 Alan Noonan 25 2.25 7.03 25 11 -7
11 Chris McCulloch 24 2.34 7.02 24 12 -1
35 Jens Erickson 21 2.64 6.93 21 13 +22
12 Peter Hernandez 24 2.31 6.93 24 14 -2
17 Quinn Wongkew 22 2.45 6.74 22 15 2
37 Chris Dyer 20 2.66 6.65 20 16 +21


18 Joshua Wilson 22 2.41 6.6275 17 +1
19 Weston Odom 22 2.39 6.57 22 18 +1
13 Colin Hanna 24 2.19 6.57 24 19 -6
14 David Hoyland 24 2.19 6.57 24 20 -6
20 Nicholas Hansen 22 2.38 6.55 22 21 -1
21 Gregor Terrill 22 2.33 6.41 22 22 -1
15 Zach Eaton 23 2.2 6.33 23 23 -8
22 Mason Hans 22 2.28 6.27 22 24 -2
23 Jason Deng 22 2.27 6.2425 25 -2
24 Evan Hill 22 2.23 6.1325 26 -2
25 Spencer Healey 22 2.22 6.105 27 -2
26 Tomasz Fedorczyk 22 2.19 6.0225 28 -2
27 Kyle James 22 2.19 6.0225 29 -2
28 Keith Gaudry-Gardner 22 2.17 5.9675 30 -2
38 Travis Yeo 20 2.38 5.95 31 +7
39 Scott Pagliaroni 20 2.38 5.95 32 +7


#2

I’d still have to play D’argenio in Double Elim first round, is there another system we can use? :wink:

I’m leary of a change that puts two Brits out of Top 16 (although another one does go back in). :smile:

Anyway, enough about Worlds, in general I think I like the idea in principle. At the big Dark Sphere SC events I went 6-0 first three games and then got really tough matches - I think I was top three for SoS, but didn’t have enough points to get to the cut - if I had the benefit of the tough matches would have been that SoS (8th place (2.08 SoS vs 2.53 that I had). Ultimately I needed to win two more games though…!

I can see the value in it - it seems “fairer” somehow, but I’m not sure how much the first game skews things and then impacts the rest of the day?

I’m not averse to the current system, but I can see the argument for a change. That said, I’d have a hard time convincing Dave H that we wasn’t in the cut when he’d got 4 more Prestige than Chris D - even though Chris was arguably battling tougher opponents all afternoon.

There’s the obvious statistical anomaly of taking the final rankings and massaging them compared to what might have been (i.e. if every round Prestige * SoS was the basis for finding your opponent).


#3

Hi Gary, thanks for the feedback - no system is perfect but I do think that if the aim of the Swiss is to find the best X players over a limited number of rounds of swiss then weighting accumulated points by their relative difficulty is the way to go.

From my own experiences (first tournament season) I have been front running in tournaments, loitering around the top tables and then bombing out in the final rounds and missing the cuts (SoS has been as high as 2.75), so maybe Im just biased. Ultimately, just “win your games” is the best remedy, but I cant help feeling a little slighted that all results are considered equal in Swiss, when clearly that’s not the case in the reality. I don’t know if it regarded as a valid swiss strategy to tactically drop a couple of early games to set up the big finish but maybe that is a viable tactic for the more tournament savvy.

FYI - I re-ran Dark sphere and under the new metric and in this alternate reality you would have placed 10th instead of 13th, as Ben Ni (with 2.44 SoS & 16pts) does enough to squeak into the cut ahead of you. You do finish above Tagore though and that has to count for something.


#4

this is just disappointing all over again


#5

after some more tinkering a more elegant way to show the same metric, instead of arbitrarily scaling by N would be
Prestige * SoS / mean (SoS)

mean (SoS) will usually be 2, or if there are some timed wins, a little bit under 2.

essentially all this is doing is adjusting your Prestige by the relative performance of the opponents you faced during the day vs the average, but it gives back a number which is still recognizable as Prestige.


#6

I feel like it’s important to maintain the feeling for players that “If I just do well enough this round, I might still have a chance.” The ability to mount a comeback from a bad start helps keep people hopeful about their tournament performance, and I think if we have a system in place that more heavily rewards players for getting to a top table early, we’ll see more round 1 drops.

Another advantage of the current system is that it’s easy to calculate, and easier for players to understand on their own. Prestige just makes sense, and nobody has to guess where they are right now.

Not saying it can’t be improved, just pointing out some of the current system’s strengths.


#7

I agree. This system seems to put less importance on winning games and more on playing good players. As in any sport, your overall W/L record should always #1 evaluator with everything else just being tie breakers.


#8

As someone who had to manually round up and calculate the top 8 at an SC (asa participant, not a TO), I 100% agree with keeping tournament rankings as simple as possible.

I would also like to support the current system that allows about 25%-50% of the field to squeak into the top 8 if they just win big in their last round. Without some chance of making it into the cut, many players drop out early, increasing the importance of strength of schedule penalizes players whose opponents drop.


#9

The way the formula is constructied the points you get still determine 90% of the order, this is simply changing a few places around where its seem plainly obvious that someone has a much rougher ride during the day.

I can see the case for simplicity, but i dont think it equates to fairness as swiss is a very small sample size trying to approximate the round robin equivalent results


#10

While I understand that I still dont think someone with a worse record should ever rank higher than someone with a record better than theirs. You cant control who you play. If the 2-0 you play second round loses the rest of their games you’re fucked and you cant control that. They only thing you can control is your wins and losses. Is this similar to the “fairness” suggestion of Regional bye winners forfeiting their bye because their SC had less people?


#11

I see what you’re saying, but if the underlying logic is correct and this system is really picking “better” winners, then the psychological component of losing SoS because a player you beat went on to lose a lot of games isn’t actually relevant. If you need to feel like you’re in control psychologically, you can always just go with “just win all your games”. There’s no way the proposed system will send you home without the plaque if you don’t lose any of your games.

Of course, Netrunner’s not a zero variance game, so regardless of the tournament structure, not every tournament experience is guaranteed to leave you feeling like you have a high level of control. I’ve kept a 1 agenda hand at a top table and topdecked 4. Prestige and SoS and all that jolly stuff was whatever.


The concept of SoS possibly being more important than record is pretty intriguing. It’s kind of like, Alice has .250 batting average against Randy Johnson (a godly pitcher), Bob has a .270 batting average against Tom Glavine (average pitcher). Is Alice better than Bob? Maybe. If it’s the only time we’ve seen Alice and Bob play, which one we say we think is better is kind of an opinion, maybe a good pitcher only ever does 1% better than an average one and Bob is clearly better. But if we have a lot of sample data where lots of Alice’s and Bob’s have faced differing pitchers and then later crisscrossed, it seems it could be possible to get a hard baseline to back up an assertion that Bob needs to bad .150 points better than Alice to get on her level.
But that .150 is a value from empirical data. Both here and with netrunner, I think it seems the weighting on SoS has to be based on empirical data. There’s no naturally arising correct way to balance SoS with total wins. You can’t even use a different game’s data, because the skill deltas will cause different win% delta.
I could have used a symmetrical metaphor, but I didn’t, by the way… Netrunner’s not symmetrical either. I don’t see a reason the same formula necessarily needs to be used for the corp side and runner side.

If you can gather enough data to prove out a formula that can predict wins and losses later that tournament more accurately than the current system, then I think that formula should indeed be used, whether or not it “feels” bad. Most people complain about conventional tiebreakers anyway.


Aside from all this, Netrunner has a lot of variable game lengths and wasted downtime, so I kinda wish tournaments were run without parallel rounds, which might make this moot (though it could still apply). I’m probably going to repeat that every time tournament structure gets brought up. Sample size is always way more powerful than clever formulas, and there’s at least one if not two additional rounds of play hiding between games of a Swiss when an Joey the NEH Crim mirror participant has to wait for a Congress Kate to lose a bunch of psi games to RP glacier before he’s allowed to go play more games and prove his first round loss was a fluke.


#12

It’s an artifact of the tournament structure. 0 or 4 points is a large swing of points. Swiss works fine; the problem is that you can lose two games in one match and it can drop you below someone that was below you before, but swept their match.

The answer is to have single-game matches. Chess has been doing this for a while; sure that’s not a great argument for everything Chess does (no. no clocks. stop.) to import over to Netrunner, but this makes sense. They’ve dealt with the problem of players starting with a different setup (white vs black), even if in Netrunner it’s not a perfect analogy (I don’t think Runner vs Corp has a stable win-rate difference, like White has in Chess.)

I’d want to try running a tournament with single-game matches and see how it feels… Especially if you also randomize Runner/Corp and give that as part of the Pairings for each round. (While trying to make sure each player plays both sides equally, and only randomizing if both are equal.)