Frogboy

i like this, but it would also be nice to have a shuffle teams option, then the ability for teams to disguise themselves as random players (what generally tends to happen) could be partially negated.

The connection dialog is Stardock so we can do that.

The intent is good, and I'm glad you're considering getting moving in that direction.

Now, as to the actual suggestion...

First, forget about experience rating.
I know you've spend a lot of work on it, but the sad truth is it's a complete failure. I hear you've based the rating system on the chess ELO rating, well chess and DG are two totally different beasts. One is a 1v1 game with slow progression, the other is a team game with fast progression (you can become a decent DG player in two week's time, with chess, try years instead).
Bottom line is the experience rating doesn't tell anything about a player's skill : you can have 1100 (good rating) and be a hopeless newb, you can have 100 (abysmal rating) and be a very good player.
Rating players ina game like  DG is a very tough problem, and I'm sorry I can only tell you've failed at it without having a suggestion for a better system.
But I do have a suggestion as far as the displayed information : total number of games - win % - win % over the last 20 games.
Not ideal, but a lot more useful.

As for the "on this team %" display...
First, can the connection overlay get data from the game lobby ? Because if you can't tell on which team the player is, the idea is stillborn.
Second, well, that's your topic title : what algorithm shall be used ?
Shall the exact team composition need to be matched or any combination of it's members ? (ie, shall the 2-player premade + 1 random guy be detected in a 3v3 game ?)
What part of the game history shall be considered ? Total games played ? Last 30 days ? Last week ?
I believe that to get a meaningful value for that "team %" would be incredibly difficult.

So my suggestion would be to change the display : on line per player, with additional information for players on the same team : total number of games played on the same team, number of games played on the same team out of the last 20 (if total number of games played > 20).

Frogboy
Mr_Wizard : connected (proxied), 31 games together, 17/20 last
Gank : connected (proxied), 6 games together, 6/6 last
AmaZinG_Hero_ON (proxied)

Mr_Wizard
Froboy : connected (proxied), 31 games together, 17/20 last
Gank : connected, 3 games together, 3/20 last
AmaZinG_Hero_ON : waiting for connection

etc...

Some redundant information, no information about the team as a whole (only about players' familiarity with one another), but I believe it would be better than trying to sum up everything in one figure.

Firstly, nice suggestion.

I only think it's hard to implement and prone to "cheating". There's always a way to "game the system", as you can see with the current disconnect/desync exploit.

Secondly:

To me, the easiest option to keep premades and PuGs apart would be a simple "shuffle players" option which can be checked when hosting a game.

All players will be assigned to a random team after the game starts, so the team in the game lobby does not count. Premades would simply not join such games because they would risk playing against each other. Moreover, the host can also not influence the strength of his team by kicking people with low stats, which happens quite often.

I have to agree, I have seen a number of incidents where ELO has been wrong in demigod.  The above seems better.

I agree about the experience rating being a bit messed up... or simply stated, I don't use it... I use games played/win loss ratio/rank

I agree with not using experience as a measurement of talent or ranking.  It is the most meaningless statistic I've ever seen in any game.  As others have suggested, use the win-loss stat.  I would also like to see an "average favor points per game" statistic for each player, since that's a truer measurement of individual talent.  Win-loss records only reveal how often you were on a winning/losing team, which is not necessarily reflective of individual talent. That is, I am more impressed by the player who scored 100 favor but whose team lost the match than the dude who earned only the minimum 75 favor for winning the match.  So, take total favor points earned and divide by total games played (each of these figures is already part of each player's online record).

On the issue of premades, I think that Brad's suggestion about showing stats on how often people have played together is a good one.  However, like others, I would also like to see a random-teams options that shuffles teams at game launch.  That should solve the issue of premade teams being in a game unannounced.  Those who don't want to be in random games or might fear that two UBs might randomly end up on the same team would still have the option of playing a regular custom game without the "shuffle" option. 

So, the stats would look something like this (IMO):

1. Wins/Losses: as actual numbers, not just a ratio or percentage (it is better to know if a person's 80% win/loss percentage is due to a 400-100 record, as opposed to a 4-1 record)

2. Average favor points per game

3. Percentage of games played on the same team

I agree that the experience rating is pretty worthless the way it is set up.  Unless you have a very high win percentage the experience rating penalizes you for playing more.  Ironic then that it is called experience.  Using favor as a rating is not a good idea though because the ability to aquire favor is not balanced among the demigods.  I do not think people should be penalized in the rating system if they play a support role.

I see your point, and I am very sympathetic to this view.  I play QoT a lot, so I know all about the tradeoff between favor points and playing a support role.  Still, the favor-point average would be one of a number of displayed ways of judging a player's experience and skill, along with wins/losses data, premade stats, etc.  And I think that you have to judge apples vs. apples (that is, you would expect a decent UB player to have more average favor than a merely decent QoT or Rook).  Overall, I think that the favor-point average could be helpful in sorting out how teams are matched in the lobby.  For example, if you have an Erebus player with a 50% win percentage who averages 105 favor per game and another Erebus player with a 75% win percentage who only averages 65 favor points per game, whose the more solid Erebus player (as opposed to the one who happens to land on a winning team more often)? 

It would be nice anyway... having to alt+tab or open a browser is silly IMO.  No prob with doing it as needed though...

I think the issue is the figures for the win/loss exp gains, personally. In order to break even against a higher ranked opponent, the ratio is something like 10 kills for every one death. Thus, the requirements to achieve higher ranks are basically to absolutely, no questions asked dominate your opponents or run away unless the fight is so heavily stacked in your favour that you can't possibly lose. I don't think it's a realistic representation of how this game plays out, based on my experience anyway.

I should probably read the whole thread next time.