Very well. Let's completely abandon Elo and anything derived from it (which includes TrueSkill and Glicko) as a possibility and start from scratch. What makes you think the matchmaker keeps track of aggregate statistics?
To be fair, there's probably some portion of it that's derrived from elo at some level. It's not in Nintendo's DNA to use any technology off-the shelf, but it's not in their DNA to invent from scratch either. But they're one of the worst "We Didn't Invent It Here Syndrome" examples. Their MO is generally to take off the shelf technology, and then re-engineer it in their own image so it works in a "nintendo-like-way". So I imagine based on the usual Nintendo procedures the system is a homebrew, but probably started with an elo base as TrueSkill and such did. That's probably why it seems to exhibit SOME traits of elo, because it's in there, but doesn't really fit with elo properly because it's hack with an attempt to make it "better" (and maybe it IS better, but is flawed.)
Since we discussed elo patterns and such, I've been paying attention to win/lose streaks, lobby exits, etc, etc. expecting to see the influence of the streaks on how it makes matches (lose a lot, and it should tank my player value, etc.) but I don't see any correlations between w/l trajectory and matchmaking difference. None at all in fact. Elo should adjust based on w/l. This doesn't seem to be using w/l in any way. Initially when we considered it it fit the pattern well enough, but I've since gone way up and way down several times, and raised an alt into the A's. The pattern of matchmaking difficulty, or changing of difficulties has not responded to wins and losses.
Raising the alt has showed a lot of trends I was seeing in the upper ranks (A/A+/S) in a clearer light without being bogged down with "maybe I'm not good enough" since I'm playing quite a few ranks below my real rank. Early in The C's I started being paired against other alts (C+'s obvious advanced playstyle, players that could outmatch me, etc.) This is before there was too much win data to seed anything. It clearly didn't use the TW 10 levels of data since my first round was against true C- scrubs, and then it very quickly ratcheted me up to face other alts. In the B's, I started losing some. And even had a partial losing streak in B, and some extreme losing streaks in B+. I treaded water and even almost went back to B several times. I was being paired against a great deal of fellow alts. Players in B, B+ with levels below 20, advanced play, etc. Even with losses, including more losses than wins, it did not adjust the difficulty of these battles or stop matching me with alts, or match me on the "advantaged" team - I always had the lower rank teammates, the other team always has the higher rank teammates and/or alts. While hearsay, it seems to be conventional wisdom within the game that alts are generally matched against other alts. Now, elo could explain that because they win more, but in these alt v alt battles, w/l would remain similar to playing within your own rank because you actually are. So your w/l wouldn't actually look very different from other B- players, for example, who are not playing the all alt lobbies.
Now in the A's, the alt account is back to the familiar pattern where my team is all A-, and the other team is all A/A+, even in A-. It clearly knows I'm the "better" player (with less play data), and pairs me in uphill challenges accordingly. Despite being very loss prone, like a real B+ during my rise to A-.
It's very clear now that W/L isn't a significant factor, if it factors W/L at all. Which is TECHNICALLY a good thing. it DOES seem to know that despite losing as much as a B+ player rising to A-, that I was playing at a higher level than a B+ others. That's also a good thing. But I think it must be overevaluating whatever "player value" it does use based on some unknown conditions. Or it overcompensates (based on limited player availability?) in how it matches you as a result. I do think it's largely going for a 50/50 avg team value, though there seems to be a bias, beyond the numbers of a player getting the lower average, if Splatfest is an example. I consistently got paired on the lower valued team, even when it was showing us our "fes power" number explicitly. I'm assuming there was still a little bit more behind the scenes in that matchmaking. Similarly in ranked, the consistency of getting lower ranked players on my team, etc, if it's really going for a 50/50 is a little startling. If it were using w/l that wouldn't happen nearly so consistently.
Overall, It seems increasingly clear that it's not using a W/L derived result as normal Elo would. It' using "something else" (and the something else would have to be aggregate statistics) which is technically a good thing. It seems like it does try to match based on a 50/50 - not chance to win, but average team value (can't be chance to win if it's not using W/L as the value condition.) That would also explain why "the other alt" is always on the other team and never on mine, and never randomly switches them to mine (but switches their bad teammates to mine) - because we'd have the similar values, and therefore would be the opposing match for team value balance.)
So what about my main account? Noticing the same type of patterning on my main account as I did coming up through the lower ranks on my alt, similar to what jsilva said, I believe there's something wrong with the aggregate statistics tracking that can cause a player to become overvalued based on some condition or play style. Some multiplier somewhere along the line that gets inflated based on some condition. I suspect, while it's not explicitly finding alts, my alt gets paired with other alts because both players have a much higher player value than other players in the rank and so it pairs them against each other. And I'm starting to realize that's what happened in my main account too. Just as I got paired against A+/S alts in B- on my alt account because I had similar performance to them (because I really am one), I believe on my main account at A+/S, I get paired against the S+ alts because it effectively thinks I am one - I must have a similar player value based on some condition of my statistics, this would in fact HAVE to be somethng other than W/L as my W/L would be quite bad compared to middling A+/S players, that those alts do, and it pairs me against them, effectively trapping me below my actual rank because the matches I'm playing actually ARE at that rank. Except in that case, I'm definitely NOT an S+. It's matchmaking for my alt was accurate in determining I'm the same as those higher tier players, because I am. But it falsely thinks, on my other account, I'm the same as the higher tier players in A+/S (which would be S+ alts) even though I'm actually not.
One side note, you probably remember me early on mentioning something seems to change at map rotation. I'm now positive of this. I've rarely seen it NOT be true. I used to believe it simply becomes HARDER after the 9:00 (now 10:00) map change. I believed maybe time of day was it. But I've seen it work in the opposite direction now too. The most stark example was buidling my alt in B+ the other day. Similar to days prior, I was getting matched with alts constantly. Gameplay was fierce. FULL A+/S level gameplay through and through, from BOTH teams. I was having a hard time getting anywhere....losing a lot...then w/l/w/l, then won a few, etc. I ended up right around where I started. I was changing lobbies a lot, and it didn't affect the performance of either team. Then map change. Suddenly after map change it became easy and B+ played exactly as B+ should be expected to play, and I easily won almost every round carrying hard, and delivering the ko myself all but twice. My player value certainly didn't change between map change. Something else changed somehow. Maybe they have "offsets" based on rotations or something that modifies values? That's just a wild stab in the dark, but I'm quite positive that rotation/mode has an effect on matchmaking beyond being a "good map or bad map" for you, and apparently, not time of day related simply because "the Japanese have arrived" etc.
While it doesn't bring us anything but more theory on what it MIGHT be doing, I'm fairly certain now that what it is NOT doing is using W/L to determine player values or assemble lobbies as a primary factor.
