Phoenix, right, rounds won is the measure of success, then you work back from that to figure out the variables. This is what Jon has done. I think a hybrid method is best, to reduce the 9 games needed to spot the outlier scout who will be singlehandedly determining results in the meantime. More obviously in tf2center than tf2pickup.

[...] Nice write up. How about doing one for your main spy? :D

w/l rate

You shit talk the player playing the class on the enemy team, and see if he's able to come up with a response, if not, then you're doing fine. No but seriously, if you're are trying to judge from the logs, as a Heavy you'l be looking at things like their Soldier/Scout/Heavy kills since the Soldier and Scout are often the ones rushing for your Medic. If you're being killed by the enemy heavy more than he is killing you, then that means you're being out positioned a lot. You also want to be focusing more on damage as a heavy, with the least deaths possible. But yeah, logs do not always show everything. I remember playing a TF2C lobby with you not long ago in Mumble. I wouldn't look too much in TF2C logs though, since most people there don't even know what they are doing.

Logs are not enough. You simply cannot score classes based on logs, just like how you cannot fully judge one's performance from their logs.

Defining stereotypical roles for classes then rating their stats based on it punishes people who play a class in an unorthodox but successful way.

you cant have a formula for that kind of things, imagine this scenario you have a scout trolling around 4 people while preventing the point to be captured he died and did jack shit damage but sucessfully stalled 4 people from capping unting his team arrives, while he didnt do any damage he was still usefull in that situation. Its like having a spy cap in every round, while you dont do damage youre still being usefull

...