For my (fellow) nerds, I wanted to address two main things.
1. Defining Plat+ (Classification Accuracy Shrinking Sample Size)
TL;DR 1: Outlined below, we currently use a classification system (A) which we believe prioritizes accuracy, but does lower the sample size pool (ex: Plat+) by a small amount. The old system (B), which most sites use, inflates win rates and sample sizes - plain and simple, it's not acceptable to use a system that says a class (Plat+) has a 55% winrate on average. That would incorrectly mean that most builds have a positive win rate in Plat+. That skews the data way more than sample size. I'd try to remember that size of sample is not the be-all-end-all of accuracy.
Let's use this sample game:
Team 1 (W): Plat 4, Plat 4, Gold 1, Gold 1, Gold 1 | Team 2 (L): Gold 1, Gold 1, Gold 1, Gold 1, Gold 1
As a simple example, here are 2 classification strategies (among many). We currently use Cla...