Ok Champ, I think I get the +/-20 shift now.

What I still cant make the final connection on is how this is fair:

1st Player2500 -1.6
2nd Me1200
3rd Player1200 +10
4th Player1200 +10
5th Player1200 +10
____
1st Me1200
2nd Player2500 +18.4
3rd Player1200 +10
4th Player1200 +10
5th Player1200 +10

28.4 vs.48.4 - I just beat the strongest player in the game, but was only awarded the 20 point "shift" to my Elo, exactly the same as I would have for beating someone equal, or even worse than me. Right? Seems like in terms of the reward, the rating of the other players only matters for establishing an average (thus, the term prepooled I was using before)

We shouldnt be rewarded the shift. We should rewarded the gains from wins, minus the loss from losses.

Champ, I'm still curious on your thoughts on this.

Am I wrong about this "shift as reward" thing?

This is correct but misses two key pieces of information - the neutral point and how the "live rating changes" are calculated.


Firstly whilst you are correct there will always be a difference of 20 between a win and a loss in a K-20 ELO the upside does not always equal the down side.

A 2500 player will only gain a couple of points for beating a 1200 but potentially looses 18-19 points for a defeat. And vice-versa for the 1200 player. So you are correct you only gain 20 points for a win but this doesn't ecver start at zero. In the case above its either -2 to +18 or -18 to +2. So the 1200 player has a much larger upside!

The second pint is that the game calculates all of this at game start and creates the points tally visible on the side. The point though is this scale is different for every player.

In your example the 2500 player would probably have to come first to get a positive score (+8) but second would be 2+2+2-18 = -12

If he came second another of the +2s becomes a -18 giving -32 (a 20 point shift)

For a 1200 in this game though there would be bonus of +48 for first (10+10+10+18) and then on downwards.

So the key point is that your initial ranking sets the tipping point between winning and losing points but each place will be twenty different at all times.

kevlargolem wrote:Ok Champ, I think I get the +/-20 shift now.

What I still cant make the final connection on is how this is fair:

1st Player2500 -1.6
2nd Me1200
3rd Player1200 +10
4th Player1200 +10
5th Player1200 +10
____
1st Me1200
2nd Player2500 +18.4
3rd Player1200 +10
4th Player1200 +10
5th Player1200 +10

28.4 vs.48.4 - I just beat the strongest player in the game, but was only awarded the 20 point "shift" to my Elo, exactly the same as I would have for beating someone equal, or even worse than me. Right? Seems like in terms of the reward, the rating of the other players only matters for establishing an average (thus, the term prepooled I was using before)

We shouldnt be rewarded the shift. We should rewarded the gains from wins, minus the loss from losses.

seethestar wrote:Firstly whilst you are correct there will always be a difference of 20 between a win and a loss in a K-20 ELO the upside does not always equal the down side.

You are right, the upside should not always equal the downside under ELO, but in Subterfuge we are in fact rewarded the Shift rather than the actual ELO score. We always gain or lose 20, regardless of who we pass. The only effect ELO has is to determine a pre-averaged neutral point.

seethestar wrote:So the key point is that your initial ranking sets the tipping point between winning and losing points but each place will be twenty different at all times.

Yes, your place will be a 20 point shift difference. The problem as I see it, is that you are actually rewarded that shift amount, not the amount you gained.

Aka if your ELO gain would be +18 and the person you beat loses -2, you are inexplicably rewarded with +20 points of shift directly to your rating, and your opponent inexplicably loses -20 points from theirs.

Subterfuge calculates the ranking change as if you played a 1v1 ELO ranked match against every opponent. So your points gain for coming first is identical to the ELO gain you would get from beating everyone in a series of 1v1 games. Play vs high ranked players and this maximum gain is large (18 X number of players) play bs people well below your rank and its tiny (2 X number of players)

Since this is a K20 system you then must gain 20 less points for everyone you lose to (regardless of their rank) - if you now lost to a 2500 player you don't gain 18 and loose 2 instead - or loose to. 1200 player you don't gain 10 but you loose 10 - hence why the drops are always 20 between placings.

kevlargolem wrote:

seethestar wrote:Firstly whilst you are correct there will always be a difference of 20 between a win and a loss in a K-20 ELO the upside does not always equal the down side.

You are right, the upside should not always equal the downside under ELO, but in Subterfuge we are in fact rewarded the Shift rather than the actual ELO score. We always gain or lose 20, regardless of who we pass. The only effect ELO has is to determine a pre-averaged neutral point.

seethestar wrote:So the key point is that your initial ranking sets the tipping point between winning and losing points but each place will be twenty different at all times.

Yes, your place will be a 20 point shift difference. The problem as I see it, is that you are actually rewarded that shift amount, not the amount you gained.

Aka if your ELO gain would be +18 and the person you beat loses -2, you are inexplicably rewarded with +20 points of shift directly to your rating, and your opponent inexplicably loses -20 points from theirs.

seethestar wrote:So your points gain for coming first is identical to the ELO gain you would get from beating everyone in a series of 1v1 games. Play vs high ranked players and this maximum gain is large (18 X number of players) play bs people well below your rank and its tiny (2 X number of players)

This much is true. But only because you're looking at game wide averages. When you look at the game as one big average (like Subterfuge does), it looks like ELO is working correctly.

The problem is that the game is not one big average. It is harder to defeat one player than it is to beat another player. Therefore rewarding based on a 20 point shift regardless of who it is makes no sense. The reason this is very important, is because under the Subterfuge system of averaging, you should ALWAYS ally with the strongest player, because just by being in the game with him, you are gaining ELO due to him helping set your neutral point. Now that the neutral point is set, you want to pick up as many 20pt shifts as you can (which come easier from the weaker players), and not worry too much about beating the strongest guy because his 20 would be much harder to earn.

seethestar wrote:Since this is a K20 system you then must gain 20 less points for everyone you lose to (regardless of their rank) - if you now lost to a 2500 player you don't gain 18 and loose 2 instead

If our Elos say I should win 18 points, I should not magically be awarded an extra 2 because the total amount of shift was 20. It should make no difference what the shift was. I should win only what I win, or lose only what I lose. Not an automatic 20 based on a predetermined neutral point.

Firstly, apologies about missing this a lot earlier. I went away for a few weeks and am catching up now.

kevlargolem wrote:If our Elos say I should win 18 points, I should not magically be awarded an extra 2 because the total amount of shift was 20. It should make no difference what the shift was. I should win only what I win, or lose only what I lose. Not an automatic 20 based on a predetermined neutral point.

You are being awarded 18 points if you beat someone ranked a lot higher than you. There is no bonus 2 points being added magically. It just appears like there is because in the live ladder if you are sitting below that high ranked player the live ladder is assuming you will lose and has subtracted the 2 rating points from your total already even though the game has not ended.

If you beat the player you get a 18 point gain but if you lose you get a 2 point loss. That's a 20 point swing, however it is heavily weighted to account for the fact that the player you are up against is ranked a lot higher than you. The illusion of everybody being worth 20 points only comes around because the live ladder is applying win/loss rating changes as if the game has ended, which it has not.

Once the game ends you will see a variety of rating movements from all the different players that are not on a 20 point pattern.

In a 1v1 chess game using the Elo system the same thing happens. In every game the difference between their rating change for a win versus their rating loss for a loss equals the kValue (different kValues apply to different players depending on current rank). You just don't see the change as being the same as the kValue because nobody looks at live rating changes as the game is progressing.

Champinoman wrote:There is no bonus 2 points being added magically. It just appears like there is because in the live ladder if you are sitting below that high ranked player the live ladder is assuming you will lose and has subtracted the 2 rating points from your total already even though the game has not ended.

Ok, I think this is the bit that wasn't clicking before. I think it finally makes sense to me. Though it might still be unfair that you should just automatically ally with the strongest player because it hurts equally to lose to him as anyone else (it doesnt hurt your score much to lose to him, and you want to prevent any low rated players from hurting you with their rating by finishing higher).

But at least in terms of mechanics/math, I think I understand it.

kevlargolem wrote:Ok, I think this is the bit that wasn't clicking before. I think it finally makes sense to me. Though it might still be unfair that you should just automatically ally with the strongest player because it hurts equally to lose to him as anyone else (it doesnt hurt your score much to lose to him, and you want to prevent any low rated players from hurting you with their rating by finishing higher).

I wouldn't say it makes the most sense to ally with the highest ranked. If you and ally with a lower ranked player and manage to beat the higher ranked one, you'll receive a larger gain than if you beat someone else correct?

v3xt wrote:If you and ally with a lower ranked player and manage to beat the higher ranked one, you'll receive a larger gain than if you beat someone else correct?

No, because the difference ("swing" as I was calling it before) between winning/losing against anyone will always be 20. Aka, if I beat the high ranked guy I'm no longer losing 2, and I'm gaining 18. If I beat the bit-low ranked guy, I'm not losing 15, and Im gaining 5.

So team with me (rating 1594), so you only lose like 2 points (albeit you lost the opportunity to gain like 18), and make sure all the other scrubs finish under you, because if any of them beat you youre losing like 18 (albeit youre only gaining like 2). <-- super exaggerated numbers to illustrate the point.

The "pre-pooled" effect I was talking about before is still happening, just not by design. Rather its because of the natural result of 1v1v1v1v1 Elo.

Correct me if I'm wrong Champ.