This project introduces a new ranking system called Uniqueness Points (UP), a product of the traditional osu! performance points (pp) and a multiplier based on the appearance count of maps amongst top players and the distribution of the scores set on the map
Appearance Count — every map's appearance count amongst the 50,000 submitted scores
i.e. Songs Compilation VI [Collab Extra] has an appearance count of 18 because
18 / 500 players have this map in their top 100.
Placement Data - a set of numbers recording where the map places in player top plays
i.e. Songs Compilation VI [Collab Extra] holds position 1, 9, 3, 1, 1,
1, 21, 38, 76, 37, 32, 42, 73, 62, 94, 24, 23, 28 for 18 players
Total Weighted PP - for each map, the sum of weighted pp taken from the 50,000 submitted scores
i.e. By adding the weighted pp of the 18 scores on Songs Compilation VI [Collab Extra], we get 10902.1084
(1) Plot the Weighted PP graph, sorted from high to low
[Show Plot]
(2) Plot the Appearance Count graph, sorted in the same order as the Weighted PP graph
[Show Plot]
(3) Fit a curve (reciprocal) to Weighted PP graph, scale it to the Appearance Count graph, and find the highest point of curvature (elbow)
[Show Plot]
(4) Compute the skewness of the Placement data (if above the elbow) i.e. From the Placement Data of Songs Compilation VI [Collab Extra], we get the skewness of 0.671
Maps below the elbow point are identified as uncontested/uncontestable, the value of these scores will receive a simple multiplier that scales with the appearance count, lowest appearance receives the maximum buff of 1.15x, while maps that are closest to the elbow point (17) receives the minimum buff of 1.05x. Increasing appearance count drastically lowers this factor.
The formula encourages unique scores to be challenged so its multiplier will be brought down
Maps that are above the elbow point can be properly evaluated by its skewness of placement data. Positive skew maps receives a nerf, negative skew maps receive a buff. The popularity index is multiplied to the skewness to give a more accurate buff/nerf. High appearance count will make the skewness more effective than low appearance count.
The formula identifies over-abused maps and punishes them, while slightliy buffing maps that are popular yet difficult for most players
Most importantly, I want to encourage attempts in submission of top plays on maps that haven't been tried before. There are 100x500=500,000 possible unique maps that could be submitted by our top players but right now we have less than 4500 unique difficulties, not even map sets, thus the algorithm heavily buff unique scores with a 1.15x if nobody else have it in their top 100 plays. Time to go map hunting!
If it isn't obvious enough, the system rewards unique skillsets a lot. With UP, finding and farming new aim slops will probably get you sniped in a few days because it doesn't take much appearance to get your multiplier down to nothing. But if you set a score on a niche skillset that nobody is able to contest? That will get a bonus for a long while, a 900 becomes a 1k!
Mappers should have less influence on the player leaderboard scene: For the longest time maps with certain skillsets are very easy to produce, and the recent ranked map changes have completely broke the balance (that never existed). The outcome of UP balances quantity mapping's contribution to farming.
Visit Rankings to explore the adjusted leaderboard, or click on a player to view their profile. The player data are refreshed every day at 3 UTC
will update this section for commonly asked questions in the future
In game:
dongdongben
Discord: dongdongben