EPTPart I – System Quirks
Here's a table of the ranking points awarded each event:
MLG Point Tiers
Pool Tiers
Overall Mid-tier Drop
Tier Drop Point Spread Pct
Tier Placement 1 2 3 4
Semifinalists
1-4 xxx 1200 1000 900 800 (100) 100%-67%
Championship Bracket Winners
5-8 - 500 700 650 600 550 ( 50) 58%-46%
Pool #2 Losers
9-12 - 210 490 460 430 400 ( 30) 41%-33%
Pool #3 Losers
13-16 - 150 340 310 280 250 ( 30) 28%-21%
Pool #4 Losers
17-20 - 130 210 200 190 180 ( 10) 18%-15%
Pool #5 Losers
21-24 - 70 140 130 120 110 ( 10) 12%-9%
Pool #6 Losers
25-28 - 50 90 80 70 60 ( 10) 8%-5%
Bracket Tiers
OB Finals Losers vs LB Finals Winners Losers
29-32 - 50 40 30 20 10 ( 10) 3%-0.8%
LB Finals Losers
33-36 - 34 6 6 6 6 ( 0) 0.5%
LB Groups Finals R2 Losers
37-44 - 4 4 4 4 4
4 4 4 4 ( 0) 0.3%
LB Groups Finals R1 Losers
45-52 - 2 2 2 2 2
2 2 2 2 ( 0) 0.2%
Championship points multiply each of these figures by 1.5.
If you prefer, here's a graphical form:
![[image loading]](http://kupax.com/files/17411_e3zf5/Picture%202.png)
Notice that the points awarded for each place drop really rapidly. This is the major factor which makes the rank inflexible. One high finish, and your only challengers are... other high finishers. If you pull down one #1 rank, and don't play at all the rest of the season, someone else needs 3 quarterfinal finishes to pull past you. This may be bad or good, depending... given the length of the season I think it tends to be less representative and weaken the final championship, especially given the (recent?) difficulty of the Open Bracket.
If you're not convinced, look at this:
![[image loading]](http://kupax.com/files/17412_vidgi/Picture%203.png)
When examining points awarded to any place as a percent of the total points given out, the curve becomes even more concave than the points chart. (For anyone who cares, a more flexible and "flat" points ranking of 52 people, such that #1 gets 52 points, #2 51, etc. gives the #1 finisher just under 3.8% of the total points and #52 about 0.07%. The top 13 players get a significantly smaller percentage in this scheme, after which the lines cross and all other players get anywhere from a little more to a lot more (the factor for #52 is around 3.5).
Another issue is that in general, I don't quite understand why the point separation within each tier of four players (beyond the medalists) is necessary. I see that pools need to be ranked for the next event. For that purpose, there are already possible systems. It seems most reasonable to me to rank by progress of opponent. E.g. for Raleigh's Pool #6 round losers (25-28), DeMuslim advanced furthest so TLO is #25. All other round winners were knocked out next round, so analyze those opponents: Hero > TriMaster > Machine, so Orly #26, Tyler #27, Drewbie #28. It's a nuisance to work out by hand, but setting up an algorithm to run results has to be easier than running the tiebreakers, right? It would also be completely logical to just rank them by match/game records. That's like falling-off-a-log easy.
Let's assume that point separation is necessary or beneficial. My guess is that it's supposed to inspire players to play their placement matches, but as far as I can tell many players still forfeit, or play the match without really trying that hard. Why is the spread so huge, especially at the top?
For those who don't visualize numbers easily, here's a graph:
![[image loading]](http://kupax.com/files/17409_bxv6w/Picture%201.png)
What this means practically is that within each tier – approximately equivalent results, even adjusting for different difficulty of opponents (e.g. inControl vs Drewbie compared to Haypro vs Tyler) – the point spread between #1 and #4 is much greater than the spread between #4 in the tier and #1 in the next tier (e.g. #4 overall is closer to #5 overall than #1 or even #2 overall). Sure, the drop from one #3 to #4 is never larger than the drop from that #4 to the next #1, but the drop from #2 in a tier to #4 is always at least as large as the drop from that #4 to the next #1 (e.g. while #7 overall to #8 overall is smaller than #8 to #9, #6 overall to #8 is the same as #8 to #9).
(While I'm complaining about things, it's also weird is that even though the final rounds of the open (losers) bracket get ranked, the points awarded are so very few as to be worth essentially nothing. You might almost say they were pointless.)
tl;dr - MLG should
1) Remove the intra-tier ranking matches, and/or significantly reduce the spread from the mean of comparable results.
2) Consider flattening the overall ranking curve to reduce the inflexibilty of the rank, making movement easier over the season to better reflect current skill. Note: I don't advocate changing anything during the season, these are thoughts for next year, which I understand they're already looking at any way.
II - Thought Experiment
Format
How many rounds does MLG have? Someone who drops into the Losers Bracket immediately will play 21 matches if he somehow makes the final. A player who starts in the pools might play as few as 8 and can't play more than 15. An Open Bracket winner would play at least 14 matches if he makes the final and could play as many as 21. The following numbers are the number of matches by a player who reaches the final from a given position:
Pool #1 from Pool, Undefeated: 8
Pool #1 from Pool, Semi Loss: 9
Pool #1 from Pool, Pool Championship Loss: 10
Pool #2 from Pool: 11
Pool #3 from Pool: 12
Pool #4 from Pool: 13
Pool #5 from Pool: 14
Pool #1 from Open, Undefeated: 14
Pool #6 from Pool: 15
Pool #1 from Open, Semi Loss: 15
Pool #1 from Open, Pool Championship Loss: 16
Pool #2 from Open: 17
Open Bracket Loser: 17
Pool #3 from Open: 18
LB from OBR5: 18
Pool #4 from Open: 19
LB from OBR4: 19
Pool #5 from Open: 20
LB from OBR3: 20
Pool #6 from Open: 21
LB from OBR2: 21
LB from OBR1: 21
While there's nothing wrong with rewarding pool play – the opponents are generally harder, after all – the current format makes it so that if you want to win an MLG and can make it into a Pool, you can finish 4th in that pool – at Raleigh, mostly 2-3 – and have it easier than any Open Bracket player. Open Bracket losers have it easier than anyone who can't win a pool from Open – TriMaster and Hero should have thrown their final Open Bracket matches.
(This isn't a new problem for Starcraft tournaments, of course. A player coming from OSL prelims will play many more matches to make the final than a player seeded into the Ro16 who squeaked past that round at 1-2 and a tiebreaker win.)
In this case, however, it's not actually hard to fix.
Step One: Start with filled pools of 6.
Step Two: Seed Open Bracket winners above #6 finishers immediately (and possibly above #5s as well: the average OB pool player finish has been 3.42 out of 6 since the pools were expanded to 6 players. At Dallas, 3.25 out of 5 (4, 4, 4, 1)). (Alternatively keep the 5-player pools and seed OB winners above the #5s immediately instead of into pools.)
Step Three: Profit! OB players still play more games than all but the worst pool players. If seeded above the #6s only, #6 and OB winners play the same number of matches. If seeded above the #5s as well, #5s and OBs play the same number of matches and #6s are "penalized" by having to play an extra match beyond that. Given the overall somewhat shaky record of OB players in pools I'm hesitant to seed above #5s, but there are also plausible arguments to be made.
Assuming MLG adopted the above procedure - for the sake of argument let's say seeding OB winners above #6s only - we get a tournament format like this:
Open (Losers) Pool
OB 1 Match 1
|------>LB 1A |
v | |
OB 2 | Match 2
|------>LB 1B |
| | |
v LB 2A |
OB 3 | Match 3
|------>LB 2B |
| | |
v LB 3A |
OB 4 | Match 4
|------>LB 3B |
| | |
| LB 4A |
OB 5 | Match 5 -------------------------
|------>LB 4B |
| | |
OB 6 LB Ro8 |
| | |
|-> OB Losers <> LB Winners |
\ \ v
\ \______________><-----------------------------Pool #6
\ |
\___________________OB Winners .-Pool #5
| |-Pool #4
| |-Pool #3
Pool #5-----------------------|-Pool #2
| | Pool #1
Pool #4__| |
| | |
Pool #3__| |
| | |
Pool #2__| |
| |
Bracket Ro4 / Pool #1 <> Pool#1
| / |
Pool #1 Loser<------/ |
| |
CB Semi _____Pool Championship
| / |
PC Loser_/ |
| |
Bracket Winner <> PC Winner
Scoring
I started with a number of assumptions.
1. First place at an MLG event is 1200 points.
2. I am going to score all participants.
3. 0-2 receives a score of 0. No bonus points for showing up!
We know the maximum possible rounds to play is 22; however since a single win, in either the first Open or first Losers bracket has the same effective result (1-2, eliminated), I chose to use 21 rounds as the basis for my scoring. Then I picked a pretty small but popular number to represent progressing one round in the Open Bracket: 8. The base score for advancing to a given round is series sum from 8*1.1564^(n-1) which gives 1200 as the #1 score. I then rounded it off, so the final result:
Last Round Won: Total Score (Finish)
0. 0-2: No points (#217-280)
1. OB R1/LB 1A: 8 (#153-216)
2. LB 1B: 17 (#121-#152)
3. OB R2/LB 2A: 28 (#89-120)
4. LB 2B: 41 (#73-#88)
5. OB R3/LB 3A: 56 (#56-72)
6. LB 3B: 72 (#49-56)
7. OB R4/LB 4A: 95 (#41-48)
8. LB 4B: 120 (#37-40)
9. OB R5/LB Ro8: 145 (#33-#36)
10. LB Winners vs OB Losers: 180 (Event #29-32)
11. OB R6/LB Winners vs Pool #6 Play-In: 216 (Event #25-28)
12. OB vs Play-In: 260 (Event #21-24)
13. vs #5: 312 (Event #17-20)
14. vs #4: 370 (Event #13-16)
15. vs #3: 440 (Event #9-12)
16. vs #2: 525 (Event #7-8)
17. Bracket Ro4A/Pool #1 Games: 625 (Event #5-6)
18. Bracket Ro4B: 735 (Event #4)
19. Bracket Final/Pool Final: 865 (Event #3)
20. MLG Semifinal: 1020 (Event #2)
21. Final: 1200 (Event winner)
The initial dropoffs are very similar to MLG's current scoring; but by the #9-12 tier the points distributed are comparable to MLG's current distribution and after that they are above MLG's scoring for all places. (Not surprising since all players are scored.) Placing would be done as suggested above, by match/game score and then opponent finish. Alternatively, the placement matches could still be held (although impractical if scoring all players). Either way, players could be simply pointed by tier (as I've done here) or ranks within tiers could be spread around that as an average (e.g. #9-12 pointed at 410, 430, 450, 470) or other scale.
The top 32 players scores in this redone format, compared to MLG's current one:
![[image loading]](http://kupax.com/files/17413_iaoef/Picture%204.png)
The above is not in any way intended to be a be all and end all of discussion on the subject. I'm open to thoughts, discussion, comments, criticisms, and (above all) corrections – I'm not infallible. In fact as much as anything this is a half-assed defense of MLG: we get a lot of complaints in the forums, but very little analysis and even fewer suggestions of fixes that might be relatively simple to implement. Of course I'm also critical – it's hard to get anything improved if we just agree with each other all the time. But these are my thoughts on the subject – take or leave it as you will.
