Pinball - tie breaking rules

[emb417]

Problem

• Throughout a season, players can have the same number of points accumulated, resulting in a tie. Currently, the visualization of ties will show players ranked higher than others if they have accumulated points in the current week of play. This is counterintuitive to how most leaderboards work where players retain their current position regardless if other players have accumulated points to tie them.

  • For Example: In week 5 of a season, Player A has accumulated 50 points through the first 4 weeks and is ranked 1st, but Player B has posted a score in week 5 and then we add the provisional points for week 5 to their season total bringing it to 50, a tie with Player A. Player A should still show ranked 1st, but the current visualization will show Player B as 1st.
  • The occurs because of how the data in processed. First step is to get the latest weeks data and accumulate provisional points based on current standings (positions). Second step is to find all other players from previous weeks and their cumulativePoints to create a complete list of players for the season. Third step is to sort all players based on their cumulativePoints, which is where players with provisional points are in the array before the other players and end up sorting ahead of other players.

• There are cases where many players are in a tie

  • First to make that many points doesn't accurately represent ranking in these types of ties
  • Using a position based metric would be a better measure of ranking

Solution Options

1. Process data week by week to sort provisional points against previous weeks, which should keep previous rankings higher if ties occur. This should also separate concerns between weekly leaderboards and season leaderboards, allowing colors to be specific to each leaderboard.
2. Use rollingAveragePositions to break ties, allowing players that rank higher week over week to be ranked higher in ties for the season points.
3. Use a sum of position percentage week over week for the season to break ties and show the player with the highest position percentage aggregate ranked higher. For example if a player's ending position for weeks are 11, 7, and 13, and the number of players per week are 25, 32, 35, then we would calculate (11/25 + 7/32 + 13/35) to get the sum of the position percentages.

Options 2 and 3 might yield the same results but currently rollingAveragePosition only exists for the weeks where players have a score.

Background chat from discord

From SmedleyBumblechuck:

Are ties listed in highest total score order?

From emb417:

For the season leaderboard, we start with the current week, which has players in order based on their scores, then we need to add all other players without a current week score, and finally calculate the season points total for all players and sort by the total points. So what happens in a tie of total points? It will show the player with the highest weekly score first. Higherpeanut and den_leaf is a good example this week; tied for the season but higherpeanut has a better score this week so they’re listed first.

From SmedleyBumblechuck:

Would you consider one where you get away from scores and instead use a sum of each week's standings to break ties instead of totalling game scores?

Like there will be 20 people who will play every week but never score in the top 10....so to compare their overall place you would sum each week's standings....( 15/38 + 12/40 + 22/45 + ...)

From emb417:

Absolutely! I’ve been iterating and evolving as I go, and this sounds like a good optimization. Scores per table are not being added up since tables have dramatically different scoring ranges, so it’s really based off of positions within the week but that’s based on the weekly score. The key is going beyond just this week and finding a way to normalize across all players and all weeks; I think your example gets there. Considering this is just rules for tie breaking, maybe the first player to achieve those points should be listed first?

From SmedleyBumblechuck:

You're asking about the edge case of if we still have a tie after doing the sum of each week's standings....( 15/38 + 12/40 + 22/45 + ...)? This is getting too complex, but I'd say maybe you look at the weeks where all players in the tie competed and sum their scores on those overlapping weeks?

[emb417]

https://en.m.wikipedia.org/wiki/Glicko_rating_system

https://www.ifpapinball.com/menu/ranking-info-2-2/

World Pinball Player Rankings Points which are awarded according to the tournament’s strength of field, number of participants, format of the tournament, and the players’ finishing position.

[emb417]

From SmedleyBumblechuck

The score approach is OK as long as the 2 people in a tie played the exact same weeks. But if they each missed 1 week, but each a different week, then how well the #1 plasyer scored that week can skew the results. Like if #1 does awesome and blows everyone away, the person is getting less credit than a week where the #1 person wasn't head and shoulders above everyone else.

[emb417]

using win percentage as secondary sort in v3.13.0