Setsul#9
I'm not sure if you didn't understand what I meant or if I don't understand what you mean.
I'm talking about completely random placement. So you get matched with a mix of far better, far worse and somewhat similarly skilled players against another random mix. The wins and losses will basically be random data. You will end up with ~50% winrate and wildly fluctuating elo.
If the system is good there should not be any long streaks, not if you get another one in the opposite direction or equal length right after that. If you were placed to low then a short streak is normal, maybe 2 or 3 losses after that if you overshoot and then pretty much 50% winrate. But if you were placed correctly, go up e.g. 400 elo and then go down 400 elo something is seriously wrong. Your chances of winning the later games were single digit percentages according to your original elo so if that were correct you would've just had a 1 in 100000 streak. Your elo must've been wrong. But if you then lose all that elo again then either you had a 1 in 100000 negative streak or your original elo was correct. If that happens regularly and/or to a fair share of the playerbase then the system is probably more random that accurate.
Assuming no random placement matches occur, if we make the assumption that a new player is low skill and the system adds them at medium skill then they will lose their first game. The system will reduce it's opinion of their skill and try again, they will still lose. This will continue until they also start to win games at which point the system will stabilise it's opinion of their skill and their results should start to even out.
This assumes that the system is not tuned to overreact to their performance and actually reaches a stable position. When they first enter the system they should definitely experience a set of similar results based on the differential between the system's assumption about their starting skill and their actual skill unless those things happen to match which won't be the case for most players.
There is also the matter of population movement. In a system where all the players start at the bottom (a season for example) different skill level players will experience change differently. The best or most dedicated (activity is always a factor in this kind of situation) should very quickly form a vanguard that separates itself from the main bulk of players, but a medium skill player will spend more time sat in the general body of players including the weakest because they are all dominated by the better players.
A regularly resetting skill ladder is an example of this kind of situation. A mature rating ladder that doesn't reset doesn't really see this problem. This is why some games that have seasons use a placement match system to separate their players so they can more quickly arrive at a skill appropriate level without this experience. In TF2 if you look beyond match results it's fairly easy to bracket players quite quickly off the back of 10 games or so.
So depending on how the placement matches system work, a game with seasons that uses them could well produce the 50/50 experience you discuss. A system that relies purely on it's rating system and has a mature ladder must make an assumption about new player skill at the start which will very likely vary from their objective skill that should produce an initial adjustment that creates a streak of results that reflect that.
Rating systems have moved on from Elo of course, it's not really used any more and dominates things like Chess for historical reasons rather than quality.
Streakiness is another consideration, and is related to random elements of the game being measured. Hearthstone is a notorious example where the best players who operate at a very high level can still lose many games in a row due purely to random events inherent in the game design, and the way that those random events can snowball into a huge advantage. Competitively configured TF2 is probably pretty stable in this regard, the vanilla version has a lot more luck involved.