Suggested Change to T11c

Edit: some of the information in this first post is incorrect, please read the solution post

T11c reads as follows:

In some cases, a Team will be asked to play an additional Qualification Match. The extra Match will be identified on the Match Schedule with an asterisk; WPs, APs, or SPs for that Qualification Match will not impact a Team’s ranking, and will not affect participation percentage for leagues
i. Teams are reminded that G1 is always in effect and Teams are expected to behave as if the additional Qualification Match impacts their rankings.

This happens fairly often in a tournament when the number of teams is not divisible by 4. This means that some teams will end up playing another qualification match that doesn’t affect their rankings at all. I feel like there are two main issues with this and both can be solved by a simple solution.

First, this gives less accurate qualification rankings. Almost everyone in the vex community can agree that more qualification matches are good as they produce more accurate rankings. However, the reason most tournaments don’t give teams more matches is because more matches means more time spent on matches. This means that some teams have less accurate rankings than they could have.

Second, there is the potential for G1 violations. Let me give a hypothetical, the 2 seed is in a position that they will move into first if they win their last match and get the AWP. They are partnered with the 1 seed for this match, but the 1 seed doesn’t have their ranking affected by this match. If the 1 seed intentionally messes up their autonomous, they will remain in first. However, if they don’t set up their robot perfectly, their auton might mess up and they can remain in 1st. I know most teams will do their best and no intentionally throw, but there are cases of G1 violations involving throwing matches.

Both of these problems can be solved by having teams have all of their qualification matches count towards their rankings. This will allow teams to have more accurate (not necessarily higher) rankings, and eliminate the risk of G1 violations. I assume the only reason this rule is in place is, because ranking used to be total WP, not the average. There are no downsides of changing this rule that I can think of, and I hope the @VEX_GDC would consider changing this.

Teams with an extra match have access to and additional AWP.

3 Likes

But it is averaged out over their total matches.

3 Likes

Agreed, not to mention the situation described is very unlikely to occur however giving a team another chance to increase their scores will occur every single time and would be way more unfair…

I see what you’re saying, but math still exists. There would have to be a calculation to renormalize the results.
tl;dr: 7/8 < 8/9

1 Like

How?

Both teams only messed up their AWP once. The team with 8/9 simply was lucky enough to have an extra opportunity to run their AWP.

3 Likes

But the 7/8 did not have the opportunity to go 8/9.

4 Likes

@Dave_Flowerday might be the best authority on this, but I do believe that the way Tournament Manager works is that there are no longer random “bye” Qualification matches. When TM moved to the WP/match and AP/match formula, I think the rules weren’t updated. I believe things work as @9MotorGang 's example - a team with 16WP/9 matches will seed higher than a team with 14 WP/8 matches.

3 Likes

True, but that doesn’t mean that the 8/9 team outright performed better just because they were given an extra match. They were able to convert their opportunity into an extra AWP; an opportunity the 7/8 team did not get.

It’s like if we were playing darts, but I was able to throw 4 darts a round while you were only able to throw 3. While we both still need skill and accuracy to do well, I’m more likely to win since I am able to score more points each round. Does that automatically make me the better player?

2 Likes

OK, so Team 10MotorGang, Team 9MotorGang, and Team 8MotorGang are at a competition with 31 teams. 10MG and 9MG are the top two robots; in fact, they’ve tied in every meaningful aspect. 8MG is also pretty good. All other robots at the event are pushbots.

Only one team at the event gets the extra T11c match, and the random scheduling gods have selected 10MG for this honor.

Somebody on the 10MG team is dating somebody on the 8MG team, and they’ve agreed to pick each other for alliance selections.

Because of the extra match anomaly, 10MG gets the first seed under your proposal, even though Team 9MotorGang has performed just as well. I don’t believe you’d be in favor of this outcome.

4 Likes

Okay, another example. This is a free-throw shooting contest.
The two competitors both shoot at a 95% rate.
Person A takes 20 attempts, making 19 of them.
Person B takes 20 attempts, making 19 of them. Person B is then given another attempt ~because reasons~ and makes it, because the player is a 95% shooter. Person A is not given the chance to tie.

Yes, Person B objectively did better. But there’s nothing fair about it.

3 Likes

While this change might produce slightly better qual rankings, the whole reason for quals is for elims seeding and picking. I might suggest you go for trying to bring Bo3 back instead of this, because it seems to me that even if you take a qual match away from each team, but you have Bo3, you’re going to have a more accurate tournament champion than Bo1 might give you. There have been many many more upsets with Bo1 than with Bo3, many of them where the better alliance loses.

4 Likes

The TM can just choose n matches out of the n+1 matches that were played. I am fairly certain that the simple random sample will produce fair results. The only downside is that the ranking will most likely be a mess before the last matches have been played.

Not quite correct on this one. Surrogate matches are still a thing as described by T11c. The WP/match, AP/match etc. change was made to make the intermediate rankings (i.e. while an event is ongoing) more representative of the actual situation. That change has no effect on the final rankings since at the end of qualification matches, each team will have the same number of matches counting towards their rankings (since surrogate matches don’t count).

7 Likes

The American Dream and most Hollywood movies are based on the underdog sometimes winning (not to be confused with UnderDog the cartoon where he always wins) I always root for the Blue Alliance during eliminations :roll_eyes:

@9motorgang re T11c - I think this is a tall lump of dirt out of a mole hill (a reduced version of Mountain out of a mole hole. )

I could get behind a feature in TM that says "This team objects to the concept of an extra match being played by them that will not advantage them, adds additional wear and tear to the robot and unnecessarily depletes a battery charge cycle, so don’t include them for a “balance the tournament out” position.

3 Likes

Only issue with that is I would think just about every team would want to be on the “no-go” list.

Ok, so I made a computer simulation to test a slightly modified version of @mtaylor 's senario. each player has a 95% chance of making their shot. Person A shoots 20 shots and Person B shoots 21 shots. The person with the higher average is the winner of that game.

In order to ensure accuracy, I ran it 1000 times. The results were as follows:
Person A accuracy: 95.145%
Person B accuracy: 94.876%
Person A games won: 367
Person B games won: 633

I would like to publicly apologize for arguing an incorrect point. I would like to say that teams should not have their additional matches count towards their rankings.

15 Likes

Possible for you to run a simulation where the percentage is calculated by randomly selecting 20 out of the 21 shots made for Person B? Curious about the result.

2 Likes

SImilar results:

Person A accuracy: 94.830%
Person B accuracy: 94.955%
Person A games won: 332
Person B games won: 668

Looks like the results are best left as is. Probability can be very counterintuitive sometime.

3 Likes

This is interesting, I seem to have received different results from you. My code is shared below. Essentially, person A makes the shot 20 times and the probability is calculated. For person B, 21 shots are made, and one of the results are discarded using another random number generator. The code is below if you would like to check it.

#include<bits/stdc++.h>
using namespace std;

mt19937 rng(time(nullptr));
mt19937 rng2(time(nullptr));

// makes a shot (95% accuracy), and returns whether the shot was made.
bool makeShot(){
    int val = rng() % 100;
    if(val < 95){
        return true;
    }
    return false;
}

// returns the accuracy of person A after making 20 shots
double personA(){
    int num = 0;
    for(int i = 0; i < 20; i++){
        num += makeShot();
    }

    return num / 20.0;
}

// returns the accuracy of Person B after making 21 shots, and randomly choosing 20 of them to report
double personB(){
    std::vector<bool> shotHistory;
    for(int i = 0; i < 21; i++){
        shotHistory.push_back(makeShot());
    }

    int shotIgnored = rng2() % 21;
    int num = 0;

    for(int i = 0; i < shotHistory.size(); i++){
        if(i == shotIgnored){
            continue;
        }
        num += shotHistory[i];
    }

    return num / 20.0;
}

int main(){
    int aWin = 0, bWin = 0;

    for(int i = 0; i < 100000; i++){
        double a = personA();
        double b = personB();
        
        if(a > b){
            aWin++;
        }
        else if(b > a){
            bWin++;
        }
    }

    cout << aWin << " " << bWin << endl;
}

According to my result, it seems like person A and person B have results that are barely different. Upon simulating 100000 matches and recording the games won, it seems like they both win for around 34.5% of the times.

This does seem to be fairly similar to the win rate for Person A for your simulation. Did you consider the case where their accuracy is the same?

Your result also seems statistically incorrect to me. Theoretically, the center of the sampling distribution shouldn’t change even if you increase sample size. The center of the distribution should still stay at 95%, but the standard deviation should decrease.

2 Likes