7 players were used on the other team. It may have been a legit scrim with a sub or the other player may have been logged by mistake when they didn't actually play, but the system has to have simple limits to be feasible. Catering for every corner case would increase the workload massively.

I dont know if it's of any use, but just in case, it happened again
http://beta.tf2playerrankings.com/TeamRank/Match/Detail/2721
(This time we won tho :p But still, the overall maths gets flawed)

I haven't given different tiers their own Elo because I don't know what it should be. If I don't record games between them I will never know. There is also no other mechanism that could leak Elo or Trueskill points into the upper tiers. I'll adjust the Elo of the various tiers when I get enough data to make a judgement. These games will continue to be recorded and processed.

I've run an update on Elo and Trueskill to reflect the tiers teams play in. Open has the lowest Elo and Trueskill, Mid's are unchanged, High has been given a boost and Prem has also been given a boost.

It's possible that these changes are a little strong, I'm hesitant to have all High teams rated above Mid teams when some of them are certainly weaker but I have to update the tier as a whole and in a comparison of overall win/loss rate between the two divisions the change is justified.

This now gives teams in lower tiers more ranking point incentives to take on teams in a higher tier.

Trueskill is a more complicated consideration because of it's uncertainty element, I've boosted those ranks but we'll have to see how that artificial change plays out over time.

nunya confirmed bottom 3 high team by Trueskill.

The (probably not really) interesting thing about Trueskill is that it's a 2 part rating system, the rating and a level of uncertainty about the rating from which you can generate a conservative rating figure (shown in the site).

Trueskill only gains confidence about a team's rating when it has both wins and losses to look at, because Nunya lose so many games Trueskill is not really sure of where the bottom of their rating really is so it's conservative estimate of their position is really low despite receiving an improvement.

By the same token 7's rating is far higher but they never lose so Trueskill isn't really sure how high their comparative skill level really is compared to the field, particularly given their relatively low volume of scrims

I know.
It was a joke.
God damn mathematicians.
You've destroyed the nunya elo memes before they even started.

You can make it up to me if you show me some stats. :D

99.9% of everything collected is output to stream or on the ranking site

Did you change anything about how elo worked today? Last night when I checked my team had just above 1100 elo and now we're at 950 suddenly, everyone's elo in open div seems to have been lowered as well.

read above?

Read the post #34 above

#40 and #41 my bad and thanks

Maybe they become more accurate on wether or not the team in question will win or loose, but the scoreline predictions always predicts the winner to get 5 rounds (or 6, in case of golden cap)

Elo has nothing to say about the number of rounds that will be scored only how close it thinks the game will be. 5 is chosen arbitrarily as a reasonable proxy value for the winning team

With the classic ESEA ruleset 5 rounds for the winning team would actually be a fairly reasonable assumption. So with that ruleset the system might actually reliably get you the score +- a round here and there obviously.

It would probably need some work on Elo to make it more responsive to the result like scaling the K factor to a log of the winning margin. 7 would be further ahead than they currently are if that was done for example.

Yeah, obviously. The more interesting conclusion is the opposite though. With windifference 5 you'd have to account for both maps and teams on top of elo, to even get a basic idea of rounds played.
I wouldn't want to do that, whereas adding scaling to k takes a few seconds and then a bit of trial and error.

So I'm not sure if "expected rounds" really adds anything to the stats. Most of the time it's going to be wrong. Winning chance seems like more than enough information.

Or do you plan on making the expected rounds more accurate? Imho it would be a bit overkill.

No I don't track the data that would even allow me to start making precise estimates of scores

7 only lost 1 game all season against champ.gg, but they were playing Stark on medic. Arctic Foxes took Elo off them a couple of times but only by getting draws

im pretty sure nR and ams team beat them in scrims at least once

Will this be a thing for s27?

^ yeah would really appreciate if this could be updated for season 27 :p

yes please

My understanding is that ETF2L are doing their own version. Adding a div has meant it needs a load of changes and I don't have time to put them in place