Thread by @NER0cs

The scoreboard never lies - but it doesn't tell the full story either

Here's a thread on stats in #CSGO, using @brendankent's metric framework 🧵

First, a quick explanation of the framework

The key thing to bear in mind is that 'factual' stats become 'proxies' depending on what we are quantifying

KPR is a 'fact' if you're measuring how many kills a player got, but it's a 'proxy' if we're defining how *good* a player is

We'll start with the purest measure of fragging: KPR

It's simplicity is a bonus but it has several blind spots that mean it's pretty rare to see it these days.

This is where Rating 2.0 comes in

Taking into account survival, damage, trades, multi-kills, clutches and openers it's the best tool we have to quantify players' performance in CS

But we shouldn't forget that it has blind spots of its own, like eco fragging and roles

There are countless measures of productivity in CS: K/D, Impact, ADR to name a few

But depending on what were 'quantifying', many can become stylistic measures

Take DPR

It's easy to think that fewer deaths = better, right? But then why does s1mple have 0.64 DPR and Jame 0.55?

It is better used to measure "style", as a more general proxy for 'passive' play taking into account how often a player saves or takes risky duels

The same can be said for Opening Kill Attempts

You'd need more context to see how *good* a player is at aggressive actions

But it's a great stylistic measure of how *often* players go for these plays. arT, for example..

Another good stylistic measure is the percentage of kills made with a Sniper

63.4% of jdm's kills were with an AWP or Scout, but only 37.7% of Zywoo's are

This is an extreme + obvious example, but stylistic measures provide important context for players we're less familiar with

69%, for example, of xsepower's kills are with a sniper

This could be due to poor rifling or just efficient economy management, but it does show at a glance how much variety there is within the 'primary AWPer' label

You may be thinking what my point is with all this

I'm getting there, stay with me - it centres on something I noticed when applying this framework:

𝗛𝗼𝘄 𝗽𝗼𝗼𝗿𝗹𝘆 𝗖𝗦:𝗚𝗢'𝘀 𝗺𝗲𝘁𝗿𝗶𝗰𝘀 𝗮𝗱𝗷𝘂𝘀𝘁 𝗳𝗼𝗿 𝗼𝗽𝗽𝗼𝗿𝘁𝘂𝗻𝗶𝘁𝘆

We make an effort, using 'kills per round' instead of raw 'kills' but still, so much context is lost.

Even things that attempt to account for this - Leetify rating, ESEA's RWS - have their own flaws: namely, their reliance on whether your team won the round

What I am after are metrics that do more:

- Measurements of opportunity directly - how many 'gunfights' did a player take? How many rounds did a player have an AWP in their inventory?
- Different algorithms for different roles

We can make a start on this by using different metrics for different roles, like I have a bit on radar charts for example

But more can be done. We need to stop comparing apples to oranges

An AWPer's 1.20 Rating is not the same as RUSH or arT posting that figure

Let's stop using the scoreboard to lie