Rating 3.0 adjustments go live

The formula for rating 3.0 has been changed to put more weight on kills.

Since the release of rating 3.0, it's fair to say there has been a culture shock. For so long, we as a community have been used to rating systems that prioritize kills, survival, and damage.

With Round Swing, we added context to those basic stats that have guided us for so long. For every kill, we look at how much it actually changed the outcome of the round. That means anti-eco kills in round where a team expects to win in 96% of occasions matter less, a positive and intuitive outcome.

But adding context can also lead to outcomes that take more time to get your head around. Finishing kills, where your team are in a 4v1 or 5v2, now matter much less than ones in 5v5s or 4v4s. Clutches have gone from having a very small impact on rating 2.1 or 2.0 to being able to swing a player's single-map rating significantly. Kills on the T side of Overpass count for more than on the CT side, because 3.0 uses map-level data for its averages.

We are not denying that this is complicated. But Counter-Strike is a complicated game — far more than K-D or ADR allows. Everything with 3.0 is based on statistical data from thousands of matches. It is perfectly reasonable that a player with an 8-8 K-D can have more impact than one with a 12-4, and we have seen scenarios like that broken down on the forums.

But it also goes without saying that neither 3.0 nor any other metric can perfectly capture the impact of every single player in every single scenario. Round Swing is an attempt to get closer to that goal, but we are still missing things like game state, the HP of players, and the location of frags on the map. It can also be skewed by single-round peaks like big 1v3 clutches or impactful multi-kills.

The goal with 3.0 was always that Round Swing acted as an impact modifier to a base of a more conventional rating. In practice, however, especially over small samples, it was too important in the final rating. A player should be punished for having a low-impact match, but active play and high output — regardless of precise impact — should still be rewarded.

Which leads us to…

What we are doing and why

The weight of Swing's sub-rating within rating 3.0 has been reduced, as has the weight of the final point of damage (or kill) within Swing itself in favor of damage share, trades, and flash assists.

Beyond that, we also adjusted the rest of the weighting. Each sub-rating was previously equal, but after undertaking a detailed analysis of both matches and events on LAN in CS2, we were pointed towards increasing the weight of kills within the formula.

That included reducing the relative weight of KAST and Multi-Kill rating, which has plenty of overlap with Kill Rating and was designed as an explosivity modifier rather than a full-blown performance indicator.

In rating 2.0, there was a 60-40 balance of output (kills, damage, Impact) versus the price players paid for that output (KAST, survival). The first version of 3.0 was slightly behind that (56-44), thanks to Swing being an even output-cost stat, but we have now restored that 60-40 balance.

Over large samples, the same players perform well. This does not sway year-long ratings much, and most players only move +/- 0.02 rating points. Even over an event, the effect is not as big as you might expect, as players regressed (or progressed) to their mean, and more passive players lost as many clutches as they won.

The goal for this change was instead to tweak single-map situations in favor of high-output performances. This does not mean a significant buff for any given role, but rather an increased importance for players to have both high-output and high-impact.

Matches like this FaZe-Liquid affair from Cologne 2025, then, will still see the scale of David "⁠frozen⁠" Čerňanský's output (1.21 eKPR, 84 eADR) be rewarded over Roland "⁠ultimate⁠" Tomkowiak's high-impact but, ultimately, losing effort (+9.10% Swing, but only 0.71 eKPR).

How Swing deals with losing players surviving

Now that clutches are rewarded more fairly via Round Swing, it's even more important that we distribute credit properly at the end of the round.

Ahead of 3.0's release, we adjusted clutch requirements to allow impactful plays that did not include kills, like ninja defuses or playing time as the bomb ticks down, to still be treated as a clutch.

That means players rightly receive a full dose of Swing even when they win a clutch without any kills in situations like ninja defuses, but also in rare, unfair, ones where the CTs might kill Ts as they exit and leave a T 'winning' a clutch. We do have a manual tool to override incorrect clutches and redistribute Swing evenly, but this is a last resort that we want to avoid.

Clip: Our clutch requirements have been stress tested by rounds like this Danil "⁠donk⁠" Kryshkovets ninja defuse.

To mitigate this, we are changing how any remaining Swing gets attributed at the end of the round, primarily affecting situations where losing players survive: When time runs out, the bomb explodes, or the bomb is defused.

Previously, the any remaining Swing at the end of the round from opponents being alive went completely to players who won a clutch, or proportionately to players who had positive Swing while the round is live (i.e., if you have a double entry and 3 CTs save, you receive the credit for winning the 5v3, like a full 5v5 to 5v0 ace).

Now, there is a more even split between different shareholders.There are four ways to get the credit for these situations: Winning a clutch (1 share), Swing from kills (2 shares), defusing the bomb (1 share), and surviving the won round (1 share).

This covers situations where clutches give too much credit and when none was previously given to distribute credit more evenly.

Tweaks to eco-adjustment

Because we were reducing the weight of Swing in the overall formula, we needed to ensure AWPers would not be underrated by the rest of the rating that has been subject to individual eco-adjustment.

This part of rating is essentially an eco-adjusted version of 2.1 with five sub-ratings (Kills, Deaths, KAST, Damage, and Multi-Kills). Because AWPers tend to have worse ADR anyway, we have adjusted eco-adjusted damage to be a bit kinder to players with the Big Green. This means that overall, the roles have a similar balance as before despite Swing losing weight, although we will continue to adjust if eco-adjustment is treating AWPers unfairly.

We have done the same in eco-adjusted KAST, although that is only a side effect of the real goal, which was to ensure eco-adjustment did not overrate the losing team in single-map scenarios.

Over an event, which rating 3.0 is optimized for, it is right that a player with worse weaponry is given a leg up in terms of eco-adjustment. But over smaller samples, eco-adjusted KAST in particular resulted in inflated ratings for the losing team, because their actions on eco or force-buy rounds counted for more.

Since KAST is designed as a measure of round-to-round consistency, eco-adjustment was disrupting that purpose over small samples. As a result, the power of eco-adjustment has been reduced significantly, and AWPers are back at the top in that sub-rating.

Eco-adjustment on scoreboards

With these changes, we have also added an eco-adjustment toggle to match pages and scoreboards, so that you can more easily see the underlying logic of rating 3.0.

'eK-eD' is the eco-adjusted kill-death score, where players get 'kill (and death) points' based on the win percentage in a duel as explained in our introductory article, and rounded to look more like a traditional K-D.

This process, where an anti-eco kill might be 0.50 eKPR, or a death with an AK against a Glock might be 1.50 eDPR, means eK-eD can go both up and down compared to K-D. Essentially, if you are favored in a duel, you are rewarded less for winning it and punished harder for losing it.

eADR uses the same process, while eKAST gives you more 'KAST points' based on how likely it is that players in that economic situation score a KAST.

While 3.0 can look strange next to unadjusted K-D scores if you've not been watching the match closely, we hope this toggle will make it a little bit more obvious why that disagreement can occur.

Effect on previous matches and top players

Previous matches have been overwritten with the hotfixed version of rating 3.0, in time for this year's Top 20 Players of the Year list. We will also be undertaking a re-evaluation of EVPs handed out before the release of rating 3.0 in July, to ensure this year's list is as accurate as possible.

As we said before, the vast majority of players do not move much over large samples with these updates. That has the exception, as always, of donk, whose incredible 1.59 Kill Rating sees him jump 0.04 points after the re-weighting process.

But we must stress that the intention of this fix is for smaller samples, to ensure that players are high in both output and in impact — not just one without the other.

Continued development

Counter-Strike is always changing, as evidenced by Valve releasing a huge economy update just as we released rating 3.0. This means that periodic updates to the averages behind the rating will always be necessary, granting opportunities to tweak the rating further at natural intervals.

While we await a large enough sample size to tweak rating to the new economy, rest assured that we will continue to work behind the scenes in the endless, if noble, pursuit of a one-number metric that accurately shows the impact of players in CS2.

Read more

Introducing Rating 3.0