My advice is to not make a ranking system. instead have med or demo captains and let them choose their teammates. you could give some stats on the prospect. including, if you can, for the roamer medic kills (or Ks + As) per life. or whatever people think would be better.

I would not play on a website with elo.

I'm not sure that a ranking system would be very helpful or accurate because stats in TF2 are only of limited usefulness. Emphasizing stats encourages and rewards selfish play. Anyone can bait their team and have a good K:D, or take all the heals and do a lot of damage. If a medic or demo gets no protection from their team, they'll have bad stats through no fault of their own. A medic could have low deaths but never be healing the right people, or running away from every fight. etc etc

You keep the stats used to pick teams invisible. Although I've just realised the OP's intent is only partly to pick balanced teams and partly for ostentation. I'm less interested in that so my comments have only been about team selection.

If anyone has a bunch of logs with known players, if you zip them up and dropbox them, I can demonstrate their ratings and the teams picked with subsets of those logs.

K/D ratios would only be good in a system if it accounted for things like a roamer dieing in exchange for popping the other teams uber and killing yourself to get a better spawn (red team on granary)

It'll never work with just one number, you need multiple. Crit-fail is closest, when the system is working they have KA:D for each class, as well as an average Elo. You could even go one higher to help balance out the deficiencies of the rating systems- KA:D, Win/Loss, and Elo. That way you know how well the person can play a specific class, Win/Loss helps remove the "was he farming ratio?", and Elo can put players into basic brackets.

Win/Loss might not be very popular, but it seems like it would help get rid of selfish gameplay that makes it appear that you're doing well. However the Elo works would probably include that but if you don't know how it works it isn't that useful if you can't use other benchmarks along with it

Teams would be picked randomly at first. After 10 games Mackey's kapm (don't use k/d) would be higher than almost all the roamers he's better than. Win/loss is useful after I don't know how many games, but much more than 10.

When I say don't use k/d and you keep talking about k/d, are you reading what I'm writing?

If the competition is irregular it's a problem. If OP is planning on having multiple concurrent lobbies that people can create and pile into, like tf2lobby, and unlike tf2pickup where they have a single current open lobby, then my method wouldn't work. Tf2lobby is shit though, so copy tf2pickup.

KA:D is an awful metric no matter what's going on. Everyone always spends so much time bitching about how people only watch their K:D instead of objectives, so it should go without saying that K:D is a bad measurement.

I don't really have any better ideas, but my advice would be don't try too hard. It's impossible to ensure that every lobby is balanced, and any attempt to would end up extremely convoluted and restrictive. W/L seems like a decent way to go. It takes longer to get accurate numbers but it's more reliable I think.

Tbh my impression of OP is as an attention seeker more than someone capable of doing something. We've had updates on tf2center for months (meanwhile Crafz made tf2pickup with no fanfare, and now everyone plays it). Prove me wrong.

it doesn't cost anything to be encouraging

Costs less to be honest. I like to mix it up.

Or maybe he just has no idea how to do this ranking system, as stated?

Also just because of the mini-discussion above, ranking by KA/D is wrong. This is a team game, and regardless - what matters at the end of a match is who won / lost. Losing team could have a better KA/D, for example.

There is more to the game than that, and it reflects in win/loss much more the in ka/d.

KA/D is definitely bad, we all agree on that.

Trath, if you zip and upload a stack of logs from your Israeli PUGs and pick 4 scouts, 4 soldiers and 2 demomen you know well enough to rate, I can demonstrate how easy it is to pick good teams. Though again, this only works for PUGs where only 1 game can be currently signed up for (which applies to most, but not tf2lobby).

If anything if they are on a winning streak put them on a high matchmaking bracket. And if they are losing a lot put them in a lower one. (dota2)

You can't use stats to determine player skill. The only thing that matters is whether or not that player wins more often than he loses. There are plenty of ways that a person can contribute to a team's victory that don't involve things that get recorded in stats.

I completely agree with #30 that HON's Elo system was great. Some people definitely did get annoyed when they'd lose a ton of ranking for failing to carry an awful team, but at least most matches overall were decently good.

I'd recommend using a class-based Elo that is dependent on whether or not your team wins, otherwise you'd end up running into issues where players could drive up their Elo on their best class, and then if they ever offclass it will result in very unbalanced games.

You don't need to determine skill, you need to pick teams. Keep stats used for ostentation separate from this.

One way to help with the win/loss ratio rating is to have it only be positive, and neutral at worst. If you end up on terrible teams over and over, it won't put you into a hole. It'll take a while for the sample sizes to get big enough to evaluate accurately, but within a few months you should start seeing a separation between the 5:0 team baiters and the 1:1 sacrifice players.

The classes aren't that important if you're not evaluating based on dick-waving stats. I'd rather have a super-smart player with terrible aim as my sniper than a dead-eye sharpshooting numbskull.

Nevermind