When two lowest-rating (or estimated rating) players happened to make a match, there was a high chance that they would be paired on the next match again.
kifuwarabe2018 with a rate of 479 and APTkai_i9_7960X with no steady rate happened to make a game
APTkai_i9_7960X won the game.
On the next match, roughly speaking, Floodgate selects pairs so as to minimize a loss function, which is the sum of rating differences between players.
APTkai_i9_7960X's rate was estimated as 679 (the previous oppenent's rate plus 200)
Floodgate gave a penalty (400) on the same pair.
Unfortunately, their rates were way lower than the others.
Even with the penalty, Floodgate made a match on them again.
APTkai_i9_7960X and kifuwarabe2018 ended up with playing 12 successive games from 2018-04-06 11:00 to 2018-04-06 16:30.
As a quick workaround, raise the same-pair penalty.
Fork
When two lowest-rating (or estimated rating) players happened to make a match, there was a high chance that they would be paired on the next match again.
There was a report on this issue: http://bleu48.hatenablog.com/entry/2018/04/05/125107. The scenario would be
APTkai_i9_7960X and kifuwarabe2018 ended up with playing 12 successive games from 2018-04-06 11:00 to 2018-04-06 16:30.
As a quick workaround, raise the same-pair penalty.