I'll add some additional comments to illustrate what I intended with my original post. It wasn't originally clear.
In the "old" Elo system the only thing that mattered was whether you were winning more games than losing. If so, you climbed and faced better opponents thus in turn giving you better opportunities to improve your own performance and thus continue to climb.
In the new system, your overall win rate is only relevant in determine if you EVENTUALLY get to your promo matches. What is more important is whether you can have a "streak" of wins at the right point in time given your current LP. For example, a streak of 3 losses followed by 3 wins when you were originally at 50 LP would (assuming equal gains and losses of LP) result in you being once again at 50 LP. However, if this same exact set of events happened when you were at 100 LP starting your promos matches, you would fail your promo series after the first two losses and be penalized 40 LP and thus end up somewhere around 80 LP after factoring in the 3 wins. You would be penalized at least 20 LP (plus opportunity cost of having done the promo series) simply because your losses just happened to occur in a promo series. It's even worse if this loss streak happens at 0 LP in a division since the system will then DEMOTE you and penalize you roughly 40 LP AND nullify your previous success in the god-forsaken promo matches. At the edges of divisions streaks are rewarded moreso than overall trends, and streaking is more vulnerable to random chance events such as disconnects, trolling, AFKs etc. This is where the system is failing.
Some argue that we should all "just get better" and then we'll climb consistently. Well, not many people in gold can perform with Diamond-level mechanics and decision making in all 5 roles given our environment of playing in gold "environments". Consequently it is unrealistic to assume we need to go 20-0 every game (even as support) to solo hard-carry the game. And in all honesty, we shouldn't have to perform at diamond or even high plat level in all roles to consistently climb.
And you are just 1 person in a game of 10 people. It requires a large sample size to determine whether that one person actually has more of an impact to their team than other people in their "bracket". However, games at the edges of divisions are not large sample sizes, they are instead 3-5 games that determine whether you climb or get penalized for not succeeding. If you fail your promos or are demoted after your promos, you are literally in a worse position than if you had not done the promo series at all.
I mentioned the disconnects because this is just one of many random factors that can influence a test that is based on a small sample size. It is extremely hard to overcome a 4v5 situation unless you are dramatically ahead.
tl;dr Promo series and games immediately proceeding promos series are broken because they arbitrarily value streak wins. These outcomes are vulnerable to low sample size issues.
