General DiscussionLots of Matches Counting that Shouldn't

http://dotabuff.com/pages/faq Read this :)

Unfortunately "game duration less than 5 minutes" and "no kills were recorded" aren't actually enough to be confident that a game didn't count. :-(

Why? As for as I know the game never counts as long as there is a DC before first blood and before 5 minutes.

Your algorithm would be more accurate if it just assumed such games didn't count and could still be wrong for whatever small exception exists 1% of the time.

I have read the FAQ, but this seems like a systematic problem - it seems like any game that has 10 heroes picked counts in DB no matter what, which isn't very accurate at all.

It seems that what you are asking for could be self-contradictory. You want games with DC before FB and before 5 minutes to count...and yet you also complain that any game with 10 heroes picked counts. Maybe something was mis-typed.

Anyway...Neither of those two complaints are accurate (In that its not 100% uniform)

http://dotabuff.com/matches/280020374
My game here I won 4v5, but it did not count because my CK quit immediately after picking his hero. This is a game with 10 heroes picked and it did not count.

http://dotabuff.com/matches/144778589
This game I won with a FB double kill very early that led to a 5 man DC at 2:25. This game did count for dotabuff stats despite the very early DC.

http://dotabuff.com/matches/279668322
I got FB double kill leading to some DC. This time the game did not count. Everyone DC by 6:42.

It is very difficult to determine without a human judgement from watching the game whether these were legitimate wins or not. What would you suggest as criterion for the game counting or not that can be done in an automated process?

"You want games with DC before FB and before 5 minutes to count"

No, I said the exact opposite. One rule that would be easy is just that, any game under 5 minutes with no deaths should never count. According to Jason there is still some exception to that. (maybe a tower falling?) but that happens so rarely the system would be more accurate to assume the game didn't count. I don't know how much information is accessible via the API, like player DC timings and kill timings, but that rule doesn't need either of those.

Ok, so games with no deaths where one team all quit before 5 min should not count in your view. What if the winning team did a level 1 roshan into a 5 man push mid and the other team gave up two towers then quit? I have done that before. I think it should count as a win.

You could decide on the timing of the first DC..or by the number of DC at a certain time...or by the kills, or by the heroes picked, or by cs, or by length players stayed after the DC, or towers.....there are so many possibilities. Sometime people abandon and then return anyway.

It is very hard to get all the games right to everyone's satisfaction...even if someone did personally watch every game replay it would still not be agreed upon by everyone which games counted and which did not.

It might help if dotabuff posted clearly the rules by which they decided games would count or not...but then if they did maybe people would just complain about how they disagreed with the criteria. Maybe its better to leave it a mystery.