4 ms·

I'd like to think I have some domain knowledge here[0]. So here's my two cents: there is no performance-based stat that will tell you if you're good. Rank is th

I'd like to think I have some domain knowledge here[0]. So here's my two cents: there is no performance-based stat that will tell you if you're good. Rank is the only thing that matters. There are however a lot stats that can tell you if you're good amongst your peers in your current rank. The idea is that once your rank converges with your skill, your stats will trend towards the median. But the good news is, most players are interested in exactly that. They want to know how they're performing compared to players of a similar rank. This was the basis for our[1] Tracker Score performance rating, which grades player performance relative to players at their level (so both an Iron and a Radiant player can have similar ratings.) [0] https://tracker.gg https://tracker.gg [1] https://tracker.gg/articles/tracker-score-our-new-performance-rating https://tracker.gg/articles/tracker-score-our-new-performanc...

jrajav 2y ago
Is this score meant to be like a derivative of one's rank? How fast one's expected to climb from their current rank? Or, more like a game-agnostic aggregate ranking?
- chmod775 2y ago
  Really only winning matters at your current rank, other metrics can be deceiving. For instance according to the tool my league account is D-tier in most metrics and gets an overall ~400/1000, except that it has a near 60% winrate over 200 games on my main role while playing against the top-5% of players in my region. As a toplaner I generally want to be on the opposite side of the map from the current objective and I'm not going to get to participate in many kills either - so my objective participation and kill participation are both D-tier. If I made improving those metrics my goal I'd be playing incorrectly and I'd likely lose more games. In a shooter a high KDA in [current rank] is going to correlate better with winrate, since killing your opponent is the main avenue to victory, however you could still imagine a hypothetical player who gets a lot of kills but makes the game so hard for the rest of their team, they lose anyways. At the end of the day, only your current winrate is a reliable metric for how fast you're going to climb from your current rank.
  - lwansbrough 2y ago
    Scoring for League presents several challenges as you mentioned, and I’m not particularly happy with the state of our League rating. In theory top should still be picking up standing objectives, which are accounted for, but yes KP will be lower for top. The goal is to use stats which can work for all roles but of course that is particularly challenging for support and top.
- lwansbrough 2y ago
  Essentially yes, it should behave like a derivative of rank if it is calibrated correctly. The normalized value is designed to make the ratings comparable across games, yes.
chatmasta 2y ago
I love this site. Great job! Nice to see you're on HN.
- lwansbrough 2y ago
  Happy to hear it!
kqr 2y ago
Author here. You're absolutely correct in that there are two levels of skill indicator. Something like a rank or rating is the gold standard for determining how good a player is against the rest of the group rated by the same system. It's the gold standard because it's verifiable: it translates to a probability of winning and this can be calibrated over time to increase faith in the rating. But by design, it's a fairly slow-moving indicator. Faster feedback is interesting for figuring out in-game decision rules or improving more effectively, and that's where these per-match ratings like the Leetify rating comes into play. They are naturally more difficult to verify because they are working on a very small, noisy sample, and since the outcome of the match (win/loss) is attributed to an entire team, it can be hard to detangle the individual contributions to that, without resorting to subjective judgment. Leetify have, from what I understand, tried to take a quantitative/verifiable approach to it, but they haven't shared any details around exactly how they verified it – if at all. I'll take a look at tracker.gg and see what else I can learn about this. Thanks for sharing!

jrajav 2y ago

Is this score meant to be like a derivative of one's rank? How fast one's expected to climb from their current rank? Or, more like a game-agnostic aggregate ranking?

chmod775 2y ago
Really only winning matters at your current rank, other metrics can be deceiving. For instance according to the tool my league account is D-tier in most metrics and gets an overall ~400/1000, except that it has a near 60% winrate over 200 games on my main role while playing against the top-5% of players in my region. As a toplaner I generally want to be on the opposite side of the map from the current objective and I'm not going to get to participate in many kills either - so my objective participation and kill participation are both D-tier. If I made improving those metrics my goal I'd be playing incorrectly and I'd likely lose more games. In a shooter a high KDA in [current rank] is going to correlate better with winrate, since killing your opponent is the main avenue to victory, however you could still imagine a hypothetical player who gets a lot of kills but makes the game so hard for the rest of their team, they lose anyways. At the end of the day, only your current winrate is a reliable metric for how fast you're going to climb from your current rank.
- lwansbrough 2y ago
  Scoring for League presents several challenges as you mentioned, and I’m not particularly happy with the state of our League rating. In theory top should still be picking up standing objectives, which are accounted for, but yes KP will be lower for top. The goal is to use stats which can work for all roles but of course that is particularly challenging for support and top.
lwansbrough 2y ago
Essentially yes, it should behave like a derivative of rank if it is calibrated correctly. The normalized value is designed to make the ratings comparable across games, yes.

chatmasta 2y ago

I love this site. Great job! Nice to see you're on HN.

lwansbrough 2y ago
Happy to hear it!

kqr 2y ago

Author here. You're absolutely correct in that there are two levels of skill indicator. Something like a rank or rating is the gold standard for determining how good a player is against the rest of the group rated by the same system. It's the gold standard because it's verifiable: it translates to a probability of winning and this can be calibrated over time to increase faith in the rating. But by design, it's a fairly slow-moving indicator. Faster feedback is interesting for figuring out in-game decision rules or improving more effectively, and that's where these per-match ratings like the Leetify rating comes into play. They are naturally more difficult to verify because they are working on a very small, noisy sample, and since the outcome of the match (win/loss) is attributed to an entire team, it can be hard to detangle the individual contributions to that, without resorting to subjective judgment. Leetify have, from what I understand, tried to take a quantitative/verifiable approach to it, but they haven't shared any details around exactly how they verified it – if at all. I'll take a look at tracker.gg and see what else I can learn about this. Thanks for sharing!