LoL player devotes college Master’s thesis to proving the dev team wrong

League of Legends streamer and notorious Shen one-trick xPetu devoted his entire Master’s thesis to proving the developers’ philosophy on balancing items wrong, explaining how item win rates actually work, and deconstructing the idea of the meta as a concept.

xPetu’s been a high ranked player for years, but it’s partially due to his signature pick, Shen, having incredible build flexibility. He’s one of the best players in the world with the Ninja, and his builds often look like troll builds compared to what most players would think is correct to buy.

Most people aren’t building Rocketbelt on Shen, yet here we are.

In other words, a Shen one-trick staked his Master’s thesis on proving he’s right about how a certain champion should be built, explained why he’s right, applied that knowledge in a way that could fit any champion, and built an entire app around it.

And yes, he graduated.

He explained the concept of inflated item win rates on characters, something that’s abundantly clear to most veteran LoL players. The most obvious example is Mejai’s, an item that’s only bought when players are already ahead in level or gold and tends to be used to snowball a match.

However, he argued this thought process should apply to a much wider net of items concerning when and why they’re purchased. He presented the idea by challenging Riot game designer Phreak’s philosophy on item balance.

“Generally speaking, if you dig in and say, buy his third most common starting item, and his win rate is actually 2% higher, that is almost always fake data,” Phreak has said when covering patch notes changes on his YouTube channel.

“Any item that is not his most built item has a default inflated win rate because the only players who are smart enough to identify this item is good are the ones buying it. They’re already inflating the win rate by being good at the champion.”

xPetu praised Phreak in his analysis but also claimed the Riot employee’s logic statement was false. Or, more accurately, that it’s “human intuition” rather than a purely data-driven conclusion. Even if his conclusion is right in the end, it may not have been the correct way to draw it.

So, xPetu sought to create something more tangible.

To provide that example, here’s a sample from the data in his 38 page thesis on Shen’s win rates taken from a sampling of matches against magical top laners based on the first item they built.

xPetu

Hollow Radiance has magical protections, making it the most commonly purchased item in those matchups as can be noticed by its much higher sample size (K) than the other options.

The data essentially shows that despite players often purchasing Hollow Radiance as their first item on Shen in magical matchups, buying it before anything else hurts its actual win rate (51.25%) compared to the estimation based on how much impact it should have (52.67% win rate) based on his algorithm.

This is a bit of a gross oversimplification of his thesis and how it works, but it paints the picture. xPetu admitted that Riot probably has more comprehensive data than he has with the limited function of his API access, but he also believes that there are valid conclusions to be drawn here even if he doesn’t have the full, second-to-second picture.

For instance, it’s pretty common for items to get balanced in how they scale with a player’s level. With this new data, the dev team could buff Hollow Radiance’s passive scaling with early levels and keep it the same later if they wanted to make it more effective for countering magical champions early on.

This would allow the devs to make decisions based on statistics rather than intuition or conclusions drawn from loosely related data. Having more concrete data would be a boon to the dev team, especially considering what a balance nightmare champions like K’Sante are. Dev time is valuable, after all.

Riot Games

Additionally, xPetu took some shots at the concept of a meta in and of itself in his thesis.

“Esports communities often share popular strategies, which can result in the overuse or misuse of some popularly recommended strategies, making the associated actions seem worse than they truly are,” he said.

It’s hard to account for players thinking they know what’s correct when, in reality, the strategy they’re using is much worse than other options.

All of this just to prove which item is best on Shen. However, this line of thinking could re-define the way players perceive the meta and how pro players go about formulating the best builds.