The Greatest Matchmaking Complaint Ever

July 31, 2013

In the brief period before TI3 begins, I thought this would be a good opportunity to interject some levity in the form of a cautionary tale.

About two weeks ago on the Dev forums, a thread was posted:

Just deleted dota2.

Today I’ve played 8 games, all have been reported in the One-sided match making.

We just played a game in which we were a stack or 4 players

2x 1200 + 1400 + 1500 wins
The MM decided to give us a 5th player with 500 wins, who fed an anti-mage and gave him a 12 min battle furry
( unwinnable game )

That was the last straw, I can’t continue to play a broken game, I get zero enjoyment from playing it cuz of the MM.

Really incapable developers … really really braindead people. Even a simple “learning algorithm” would do a much better job than what is currently implemented.

Praise the lord every day that you’ve been born in the US, cuz I can assure you that with this quality of product in eastern Europe you would get fired in the 2nd week.

2 years you can’t produce a working solution … 2 years …

Of course there are calls for the OP to provide the match ID of the game so that everyone can see this trainwreck of a game that finally caused him to break down and give up on Dota2 forever.  And he, of course, does not.  Everything would have ended there if it were not for one intrepid forum poster who dug up the OP’s Dotabuff profile using his post history.  As it turns out, the actual match replay had a very different story to tell.

It’s this Match:

http://dotabuff.com/matches/247326904

The OP is WD
The solo guy is Clinkz

A brief summary:

-1:00 Clinkz goes solo offlane. Buys his own pull camp wards.
0:30 Clinkz is up against AM, Distro and Dazzle
0:36 Clinkz’s pull camp ward is dewarded immediately
1:12 Clinkz’s anti-gank ward is dewarded.
2:10 AM takes FB from Clinkz. It was avoidable, but unfortunate. He went toe-to-toe with Dazzle to try to get into the XP zone, but didn’t manage to Skeleton Walk in time, before AM Blinked on him for the last hit.
4:59 After being zoned out, then chased from his tower lvl 3 Clinkz manages to deny his offlane tower. No one came to help defend.
8:15 Lvl 4 Clinkz dies to Glimpse, Kinetic Field, Sentry + AM ult.
9:01 WD, SD and SF come to help with offlane.
9:31 SD Kills AM
13:20 AM gets BF
16:15 WD gets isolated from his team on 30% HP and low mana. Seems AFK. AM tps on him and kills him.
17:18 WD gets isolated, Shackled and killed by the entire enemy team apart from AM.
22:05 WD rages in all chat with the same complaints as in the OP
22:40 WD rage quits on a 12-9 score.

and here’s the best bit

34:46 WD’s team wins 4v5, without him!

Score 33-12

Clinkz 6/3/4
AM 2/5/3
WD 0/2/5

Clinkz has the 2nd highest hero damage, on the winning team, higher than the DK, the farming carry who was protected by the trilane. 2nd only to SF, the mid hero.

It’s a must watch. Very entertaining and deeply ironic.

I’m pleased that I went to the trouble to find it and it’s a great lesson about rage quitting and the irrelevance of win counts.

I get that matchmaking might be a little frustrating right now.  It’s entirely possible that problems that went unnoticed during beta are suddenly much more prominent as thousands of brand new accounts are being added to the system daily with the staggered official release.  But if the breaking point for your ability to put up with matchmaking is a game so lopsided in your favor that your teammates can win it 4v5 without you, what conceivable set of matches would you actually find acceptable?

Among the many possible morals for this story, my favorite is this: if you find yourself in games that tend to be stomps one way or the other, perhaps the culprit is you.  You could actually be right that matchmaking is giving matches that are below your mechanical skill level because you so regularly throw winnable games at the first sign of adversity that you actively lower your MMR from where it would be otherwise.  After all, it’s not matchmaking’s responsibility to determine that you’re a giant baby and coddle you accordingly.  It just sees you as a terrible player who happens to get lucky, and maybe that’s not really too far from the truth.

Advertisements

Challenge Accepted: NaVi vs Kaipi Game 3

July 31, 2013

“I challenge somebody to decode the strategy of this game.”                 – TobiWan (http://youtu.be/Ohj26fppwm0?t=33m10s)

In game 3 of the Grand Finals of the Defense, NaVi blew a lot of people’s minds with a set of completely off the wall hero picks. But these selections were not the random inputs of a madman; they all fit a very specific strategic purpose.  First, let’s look at the team composition:

NaVivsKP

(Courtesy of datDota where you can find the full stat breakdown of the match)

The picks only tell half the story. NaVi ran Venomancer as the farmer in the safe lane with roaming Alchemist support, they had Mirana jungling ancients at level 1, and they had Ursa in the jungling everywhere else. These lanes might sound crazy, but NaVi had a very specific goal and these lanes provided them with what they needed to achieve it.

Let’s start with the Zeus pick. Zeus isn’t a popular pickup, but he specializes in putting out a lot of damage during the midgame when HP pools are still relatively low and BKBs haven’t been built. To get the most out of the Zeus pick, the rest of NaVi’s lineup is designed to spike in power around the same time frame as Zeus and the most apparent example is NaVi’s use of the farming Venomancer. Of course Alchemist gets more out of gold than Venomancer, but the goal of the lane wasn’t gold; it was levels. Once Zeus and Venomancer both hit 6 (and eventually 11), Kaipi has to deal with a tremendous amount of damage in a very large area of effect. This puts them at a big disadvantage in early teamfights until they have the items and levels to mitigate the impact of NaVi’s ultimates.

The rest of NaVi’s picks fill a similar theme. Mirana wasn’t going to get rich off of ancients, but she got enough, and once the enemy Bane and Keeper of the Light were spotted on the opposite side of the map it was completely safe to use the offlane for solo exp. Similarly, Ursa also wasn’t going to get rich jungling, but he would find levels and levels are all that he really needs. Ursa is the biggest right click threat in low gold environments, making him the perfect fit for this comp.

Kaipi was now presented with a problem: how do you lane against an enemy that doesn’t really have lanes? 3/5 of NaVi’s lineup is never visible on the minimap. You can go hunt them down, but if you’re not successful your supports will be underleveled by the time the Zeus/Veno combo comes online. Kaipi makes the decision to pressure Venomancer instead. They get first blood off this, so it doesn’t go badly for them. But they don’t get too much more out of it, and in the meanwhile Zeus, Mirana, and Ursa are all farming unopposed.

Once NaVi’s four cores all hit level 6, things really start to fall apart for Kaipi. When they try to farm, Mirana is free to go into full guerrilla fighter mode by using her ult and arrows to set up kills all over the map. If they try to teamfight they’re going headfirst into Zeus and Veno ults with underleveled supports. Meanwhile Ursa is somewhere, which means that someone on Kaipi’s side has to periodically check Roshan and fall even more behind in levels. This paranoia was aided by KuroKy calling out Roshan’s name in public chat throughout the match.

Immediately after Venomancer completes his Mekansm, NaVi attempts to sneak a Roshan. Ursa gets caught, but Venomancer and Mirana clean up the teamfight and an Ursa buyback lands them the Aegis. At this point NaVi commits to push as five. If Kaipi respects them, Mirana is free to go arrow fishing. If QoP or Clockwerk jump in, they’ll get blown up by NaVi’s nukes + Ursa. If they fight as 5, they’re going headfirst into over 1100 points of AoE damage from Zeus and Veno alone.  And none of this even touches on the hero specific counters NaVi has available like using Mirana’s Sacred Arrow to punish Cogs attempts, Ursa’s ability to tear through Lone Druid’s bear, and the myriad of options they had to interrupt Fiend’s Grip.

Essentially, NaVi put on a clinic for how to win solely through a level advantage, but they were also exploiting holes in Kaipi’s draft. Lone Druid works best as a complementary carry to another right click threat, and their lineup featured a lot of gank setup in Entangle, Nightmare/Fiend’s Grip, and Hookshot+Cogs but lacked the damage to capitalize off of it. Keeper of the Light is also a questionable first pick at this point and struggled when forced into roaming support mode. Had Kaipi had a second, BKB-sporting carry (from the few remaining that NaVi didn’t ban out), they might have been able to outfarm NaVi’s lineup while played defensively on their half of the map. That being said, NaVi likely expected Kaipi to pick a standard mid in response to their Zeus pick, fully locking Kaipi in to their vulnerable lineup.

Now NaVi probably could have exploited Kaipi’s draft in a more straightforward fashion, so why take a risk with such an unorthodox strat?

First, it’s not as risky as it seems because Kaipi didn’t know what to expect, and really, who can blame them? Second, SyndereN made the excellent point that losing a match like that sticks in the back of your head, and it’s not something you want hanging over you as you face elimination in the following match. Finally, with TI3 coming up, it sends a message to all the other teams that NaVi is capable of pulling out some crazy strats, and that will give opposing drafters an extra concern when choosing their selections.

As an aside, it’s also been pointed out to me that the game in question was played on the US East server.  Could the decision to run a risky draft been influenced by the server location?  Possibly.  NaVi’s selections do look relatively resilient to latency issues, so the theory is at least plausible.


6.78’s Huskar Rework Might Be a Bit Overpowered

July 22, 2013

Back in 6.78 Win Rate Shifts: First Impressions, one of the hot topics was how Huskar’s rework in the patch gave a significant boost to his public win rate.  There have been numerous charges of the hero being overpowered, and at the time I pointed out that there wasn’t enough evidence to support it.  He certainly saw a boost, but his overall win rate was still around 51%, which is entirely reasonable.  I also made the point that this kind of aggregate analysis often hides the whole story, and that it’s mostly there to identify the big trends that are worth looking into.  Now that I have a completed sample of Huskar matches, it’s looking pretty likely that the new Huskar is quite strong, but this strength is being concealed because most players have not adapted to the changes.

First as a quick reference, let’s look into how the patch changed Huskar’s skillset.

Q – Inner Vitality: Scaling Single Target Heal

Untouched

W – Burning Spear: Magical DoT Orb Effect

Buffed from 4 DPS per level to 5 DPS per level

E – Berserker’s Blood: Stacking Magic Resistance and Attack Speed Passive per HP Missing

Previously, this ability gave Huskar damage and attack speed.

The patch removed the damage and gave him more attack speed per stack.  It’s unclear whether this portion is a nerf or a buff.

The patch also gave him 4/5/6/7% Magic Resistance for every stack.  This is a huge survivability buff independent of any changes to his damage scaling.

Huskar also received 3 base strength as part of the patch.

We know that these changes were in net a buff due to the sizable increase he saw in overall win rate.  The question that remained was whether certain subsets of Huskar players were benefiting more than others.  As it turns out, the aggregate data dramatically understated the win rate increase Huskar has seen in high level play.

[Huskar]WinRateByPatch

Due to my reliance on sampling there is always the concern for natural sampling error.  There is, however, no doubt that old Huskar was a hero who struggled in the higher skill brackets and that new Huskar is a hero that thrives in them.  As a historical comparison, the +4% win rate shift from Normal to Very High in 6.78 would place Huskar well within the top 10 of  The Heroes With the Greatest Improvement in High Skill Games that I did for 6.77 as a whole.  The question then is what is it that Very High Huskar players are doing differently than Normal players?  The first possibility is that Very High have been quicker to adapt their skill builds, and there’s some evidence for this.

[6.78]HuskarBuilds

Players in Very High have been much more likely to work Burning Spear into their builds in light of the buff and the loss of damage from Blood.  For reference, only 10% of Very High players used Primary W builds in 6.77 and that is now up to 45%.  Doing the same comparison for Normal we have 14%->28%.  So how are these Spear heavy builds working out for Huskar?

[6.78]HuskarBuildsWinRate

As you can easily see, once we move up to Very High the dominant skill builds become W->E and E->W.  Everything else either has a dismal performance or of too small of a population size to be noteworthy.  Looking at it by point provides an even more stark illustration.  45% of Very High players put 0 points into Inner Vitality(Q) by level 8, and those players have a 61% win rate.  45% of Very High players also put 4 points into Burning Spear(W) by level 8, and those players have a 60% win rate.  There is very little evidence in support of anything but a 0/4/4/1 build by level 9, and in all likelihood Burning Spear should be maxed by 7.

I’m unconvinced though that this is the whole story.  One other avenue I’ve looked into is differences between item builds within the two brackets, and by far the biggest difference between the two brackets is that Very High loves Ghost Scepter For those of you that are unaware, the new Huskar is in a unique position in that his kit cancels out the two big negatives of the Ghost Scepter.  As before, he can manually cast Burning Spears to get around the prohibition against autoattacks while in ethereal form, and now in 6.78 his scaling Magic Resist cancels out much of the additional incoming magic damage during the active.  As a result, Huskar is in the peculiar position of being the only carry for which this item is indisputably core.  There are certainly other orb using carries like Outworld Destroyer and Drow Ranger, but for them the additional magical damage is a significant deterrent, and they usually have more pressing item demands on top of that.

The difference in Ghost Scepter adoption rates is quite stark.  Only 2% of Normal players picked up the item compared to over 27% of Very High players.  In Very High, Huskar players who picked up a Ghost Scepter had a 64% chance to win, which is ~8% higher than Huskar’s native win rate in the bracket and actually quite a sizeable return on investment for an item that only costs 1600 gold.  As a point of comparison, Drum of Endurance costs 1750 gold and only has an observed win rate of 63%.  Black King Bar, a (possibly former) staple of Huskar, costs 3900 gold and yet it’s observed win rate is only 64.5%.

On a side note, one other noteworthy usage shift from Normal to Very High is that Very High players are now much more likely to build Phase Boots.  Phase Boots appear to be a very viable alternative to Power Treads now that Huskar’s passive is purely attack speed.

Another angle we can take is to look at the change of item usage in Very High between 6.77 and 6.78

[Huskar]ItemsByPatch

Most of the big usage shifts are unsurprising.  Ghost Scepter and Phase Boots we’ve already talked about.  HotD is significantly less popular given how important Burning Spears is to 6.78 Huskar’s DPS.  Black King Bar also appears to be much less essential given Huskar’s innate magic resist scaling.

But what’s really noteworthy to me is that the items with the biggest win rate surges are the cheap ones.  Ghost Scepter is a bit of an exception here, but that’s only because its 76% win rate in 6.77 was due to it being a gimmick pickup in 1% of all games.  It appears that the new Huskar is much more oriented towards picking up a few cheap items for early aggression and much less of a hard carry.  This is supported by the makeshift Farm Dependency test I ran on his 6.78 data.  In 6.77 Huskar came in 4th.  Using his 6.78 data in the same list he would now come in at 27th.

So to sum up 6.78 Huskar:

  1. 0/4/2/1 into 0/4/4/1 is the safest build, and only relatively minor variations appear competitive.
  2. Cheap items are the way to go.  Ghost Scepter is really strong.  Drum appears solid.  Phase Boots are at the very least competitive with Treads.
  3. Farm/Aggression profile is now much closer to someone like a Viper.  This means an emphasis on early ganking that transitions into map control over a more passive farm-oriented strategy.

Valve Employee Comments on Matchmaking

July 19, 2013

Possibly old news for many of you, but a new Valve red name has shown up on the Dota 2 Dev Forum answering questions about how matches are created.  You should be able to find his entire post history here.  For the most part it confirms things we’ve already suspected, but let’s go over the more noteworthy revelations.

http://dev.dota2.com/showthread.php?t=70322&page=59&p=663906#post663906

>Match ID: 246510650

All 10 players in this game were in the 93rd-94th percentile MMR range. The difference in Elo’s between the highest and lowest player was 50 Elo points. The “noob” with only 13 wins actually had the highest Elo of any player in the match. (This was a smurf.) They did play poorly in this match, but in the previous match (246456658), in which they played against several 4000 Elo players, they had 13 kills and 1 death.

First we have the basic confirmation that matchmaking uses Elo (sidestepping the semantic debate on whether it’s “real” Elo or just similar to Elo).  Also of notice here, the match in question is rated Very High in-game.  This suggests that the player percentile rating necessary to be in Very High is larger than what I’ve found the match percentile to be.  There’s a lot of possible explanations for this, such as a relative activity levels in different percentiles, but no real way to test any of the explanations.  Regardless, it appears that player percentiles as low as 93rd will queue into games that are rated Very High.  Finally, this answer also confirms that Valve does attempt to identify smurfs…

>Match ID: 246567215

The 14-win player did have the highest MMR in the game. However, it probably wasn’t really high enough to push him into smurf territory. So this was not a great match. There was a player who had waited 5 minutes, which is why a match of this relatively low quality was accepted. Thanks for reporting this. It’s very helpful.

…and that this smurf detection is still undergoing tuning.

http://dev.dota2.com/showthread.php?t=98317&p=663920#post663920

Just to clarify, there was never any “win rate” calculation. Ever. It is true that a goal of matchmaking is to make even teams, so that the odds of the Radiant winning any given match is 50%. The matchmaker also will raise your Elo and try to put you in players of equivalent skill, which indirectly tries to get the win rate to 50%. However, it has never looked at your historical win rate and, for example, put you in a game where it knew that you were expected to lose, to end a winning streak, or given you a stomp to end a losing streak.

Just another confirmation that players tending towards 50% win rates is an indirect result of the goals of the matchmaking system (create 1) even matches 2) among players of a similar skill level) and never something that the system strives for directly.

http://dev.dota2.com/showthread.php?t=98284&p=663957#post663957

>246572710

The game looks pretty balanced to me. The Elos are all relatively close. Here are the Elo’s on the two teams:

R D
3172 3047
2918 2958
2810 2804
2788 2720
2409 2508

Now it’s a pretty big spread between 2409 and 3172, and it is a legitimate question to ask why in the world would we put people together with that big of a skill differential. The answer is that we didn’t. The Radiant had a 4-stack which covered that range (the highest and lowest Elos on the Radiant were in the same party). We matched them with two 2-stacks on the Dire. One of the two-stacks had the highest and lowest Elos for the Dire, and the other two stack was in 2 of the middle slots. You were a single who was also in the middle, and there was also a single on the Radiant, also in the middle of the Elo range.

So, the average Elos of all the parties were pretty close. And, player-by-player, each team had somebody on the opposing team of roughly equal skill. Given that 4 stack with the big skill spread, I think it’s hard to come up with a better way to get them into a game.

Many of the Dota2 matches with a wide spread in skill are created to cope with pre-made groups that happen to have a wide skill spread.  The only possible alternatives would be to either intentionally give these teams bad games or to disallow teams of this skill spread from queuing entirely.  If you happen to prefer the latter, you can always enable the solo queue only option.

Also of note, in this game the Radiant team was 4+1 with an average rating of 2819.4.  The Dire team was 2+2+1 with an average rating of 2807.4.  So in this case, the larger premade group did not receive much of a handicap and still got stomped.

http://dev.dota2.com/showthread.php?t=98311&page=3&p=664114#post664114

Just one last comment on this. Elo is a TERRIBLE way to give players a sense of “progress.” Many (most?) people reach a plateau, and their Elo stabilizes. It is simply not mathematically possible for Elo to keep increasing in general for players indefinitely as more and more games are played.

Given this reality, if players used Elo to measure “progress”, we would constantly be reminding them that they are NOT making any. That would be really bad.

There are people out there who won’t like this answer, but many of these complainers also think that plateauing is something that exclusively applies to other people.

http://dev.dota2.com/showthread.php?t=98311&page=2&p=663966#post663966

“Noob” is a relative term. We don’t consider a person with 150 games to be a “noob”. We have some good data that by 75 wins (approx 150 games), Elo is pretty accurate, and so we rely on it almost exclusively at around that point. If you are getting matched with those players, it should be because your Elo is approximately the same. Parties can complicate things considerably. I might be able to provide further insight into why it thought the match would be a good one if you provide a MatchID.

Finally, we have this post that states that the matchmaking system relies on Elo almost exclusively for player evaluation once the account has around 150 games played.  I find this one fun because 150 was the number I used way back in my matchmaking FAQ.  I guess if you write enough you eventually nail something.

But anyway, this summary is biased towards the pieces I find most interesting, and that judgment is often at odds with the populace as a whole, so feel free to read his posts in entirety and reach your own conclusions.

On a sad note, the other post I wanted to run with this week has to be delayed.  My script is choking on the last bit of essential data, and I have to suspect that the load of the Steam sale might be the culprit.  We’ll have to see if things are going a bit more smoothly early next week.


Abaddon Is Not a Carry

July 16, 2013

Abaddon came out late last week, and so far he’s maintained a public win rate just a tad under 57%.  It’s still too early for me to have any actual data on him, but I thought I’d take some time to approach things from a more theoretical angle and examine his kit in the context of Farm Dependency.

The idea behind farm dependency is pretty simple.  Every hero in the game scales with items, but due to the nature of the hero (primarily their base stats and their kit) some heroes scale better than others.  On the flip end of the spectrum, some heroes can have no items and still be influential at all points in the game.  The most farm dependent heroes are those with with the strongest scaling and the weakest itemless contribution.  These heroes are almost exclusively what we call carries.  While it’s too early for me to have any stats on him, it’s extremely unlikely that Abaddon will turn out to be a farm dependent hero.

The reason for this is that Abaddon lacks the most essential feature of a farm dependent carry: the attack damage steroid.  The hallmark of Dota’s item system is that attack damage items provide the only real source of explosive scaling.  Magic damage can only be amplified by a tiny selection of relatively niche item pickups.  Many of the strongest defensive items are purchased primarily for the utility of their use effect and not for any kind of stat scaling (Force Staff, Ghost Scepter, Black King Bar, Scythe of Vyse…).  The remaining defensive items are primarily intended as complements to an otherwise aggressive carry-oriented item build.  You might be able to win disorganized pub games using a Vanguard/Blademail/Heart/Bloodstone build, but if the other team has a competent carry that’s keeping up with your farm you’re going to lose.  Dota as a rule just doesn’t have hero designs that can build purely defensively and still be a constant damage threat thirty minutes into the game.

If you look at the top 20 heroes on my farm dependency chart, you’ll notice that every single one of them has at least one ability that amplifies their right click damage.  Doom is the weirdest, but I guess wolf aura counts.  Of the heroes without an attack damage steroid, the highest placement goes to Timbersaw at 41, and Timbersaw is similar to Storm Spirit in that he’s very good at converting excess mana regen into extra damage.

These steroids are important because they amplify the contribution of the items that you pick up.  Suppose you give Drums, BKB, Manta Style, and Butterfly to both Luna and Nyx Assassin.  Luna’s going to get way more out of those items because she gets 38 free damage from Lunar Blessing and extra bounce damage from her Moon Glaive.  On top of this she has more attack range, receives more agility per level, and has a significantly higher move speed.

So what does Abaddon have going for him in terms of steroids?  Well, Curse of Avernus offers 40 bonus attack speed, but if that alone is sufficient carry potential we’d see a lot more carry Beastmasters, since his aura also offers 40 attack speed.  We could then go on to point out that Abaddon also has really low agility growth, offsetting this bonus attack speed and that the four most farm dependent strength heroes all have larger attack speed steroids (Lifestealer and Huskar) or abilities that temporarily lower their base attack time (Alchemist and Lycan), but let’s just leave it at the simple fact that Abaddon isn’t a right click prodigy.

You might then argue that this weakness can be made up by the sheer amount of survivability that Abaddon gains through Aphotic Shield and Borrowed Time.  Well, that’s great and all, and I’m sure in pub matches you can find plenty of opportunities to blow someone up 1v1 when they forget that Borrowed Time exists.  The problem is  that if we’re back in a scenario where you’ve had free-farm as Abaddon, and your opponent has had free farm as an actual carry, all the other team has to do is  a) ignore you, b) blow up your supports faster than you can blow of their supports, and c) focus you down once you’re alone, outnumbered, and no longer a threat.  This is the Skeleton King paradox where Reincarnation makes him a terror against low-level, disorganized pubs, but where nothing about his kit makes him threatening enough to not ignore when compared to more competitive carries.

If anything, Borrowed Time is most threatening as a support ability.  It’s like a free, souped-up version of Ghost Scepter.  Teams like to pick off the squishy supports first, but Abaddon makes an unappealing target.  So you target the other support, but Abaddon is free to spam low cooldown shields and heals on that target and use items like Force Staff to help peel for them.  Basically, Borrowed Time makes Abaddon really effective as an anti-initiation specialist because it forces the opposing team to make a much more difficult choice on how to start the fight.

This certainly doesn’t mean that Abaddon cannot win pub games as a farmer, in part because winning a pub game is a relatively low hurdle.  What you should keep in mind is that Abaddon isn’t going to win many games by strictly outfarming the opposing team.  Instead, he would behave much more like a semi-carry in that he wants to quickly get a couple of efficient damage items and his ult and then dominate the mid-game through constant aggression.  Is this a viable pub strategy?  Given his near 57% win rate, probably, but this doesn’t make him an actual carry.  Spiritbreaker can do the same thing, but it works for both heroes by leveraging their early game advantages to cut down on the length of the laning phase for both teams.

Could a 1 position Abaddon work reliably in professional games?  I have my doubts, but even if it does, this still wouldn’t make him a carry.  I would expect a team built around a 1 position Abaddon to be exclusively focused on winning the game in under thirty minutes, and their success would be much more hinged around disrupting their opponents early farm than it would be about maximizing Abaddon’s.  Like the pub strategy, they would grab quick, efficient damage items and use them and Borrowed Time to force early clashes while the opposing carry is still ramping up.  This still doesn’t make Abaddon a carry any more than the early farming trilane Treant strats made Treant a carry.  Instead, these teams are foregoing traditional carry potential in an attempt to utterly dominate the early game, which would make carry potential a non-factor.


[Skill Build Analysis] Slark

July 11, 2013

6.78 stuff will be coming in shortly, but while I put those samples together I thought I’d do a write up on one of the last samples I grabbed in 6.77.  It’s a bit outdated, but Slark didn’t change much in the patch transition, so it should still be applicable.

First, the basic ability information.

Q: Dark Pact — Point Blank AoE nuke that both damages and purges debuffs off self.

Scaling: Damage and Self Damage go up.  Cooldown and Mana Cost go down.

W: Pounce — Slark leaps ahead.  First enemy hero hit takes damage and is leashed to the area around the impact.

Scaling: Damage goes up.  Cooldown goes down.

E: Essence Shift — Slark’s attacks steal 1 point of stats from enemy heroes and give them to Slark as agility for a set duration.

Scaling: Duration goes up.

So here’s how the builds break down.

[Slark]Builds

Pounce-> Pact is the most common build overall and sees the largest increase at higher skill levels.  Pact-> Pounce is somewhat less common but relatively popular at all levels.  Pounce-> Essence Shift is actually more common than both of these builds at lower levels of play, but drops off in popularity at higher levels of play.

In terms of win rate, Pact-> Pounce has the strongest performance at lower levels of play, but the Pounce primary builds overtake it at higher levels.

You might notice that all of Slark’s builds go down in win rate at higher skill levels.  Partially this is just a reflection that Slark himself does seem to perform better at lower levels of play, but part of it is the usual pattern of marginal builds becoming less popular at higher skill levels.  It might be tempting to interpret this as indicating that Slark absolutely dominates low level play so long as you choose the right build, but it’s just as likely that (relatively) mediocre Normal players are more likely to choose an obviously bad build than (relatively) mediocre Very High players.

In any case, what we can conclude is that

  1. Both Dark Pact and Pounce builds look viable.
  2. Essence Shift is a solid skill, but should never be maxed first.
  3. Pounce should never be maxed last.

Moving beyond that, it seems possible to me that Dark Pact builds work relatively more often at low levels of play, and I was wondering why this is.  One possible explanation is some kind of confounding factor.  Suppose that in the Very High bracket Slarks who go mid tend to max Dark Pact first, and Slarks that go mid have a lower win rate than sidelaning Slarks.  This relationship would make Dark Pact potentially look worse than it really is.  (It also wouldn’t prove that mid Slark is worse than sidelane Slark.  It could potentially be more situational, and the players are just doing a poor job of picking him into bad lane matchups.)

But suppose that Dark Pact builds really do better at lower levels of play.  One possible answer for why this happens is that Slark does farm noticeably better when he maxes Dark Pact first.

[Slark]BuildStats

Pact builds have a CS advantage at all skill levels.  This advantage shrinks some as you get higher, and it also becomes smaller relative to the baseline CS expectation.  This could suggest that in a game full of mediocre farmers (including yourself), maxing Dark Pact first could be enough to give you a dominating farm advantage.  Then as you graduate to higher levels of play this advantage becomes worth less while your competition is more capable of using Pounce effectively to create early kills without giving up as much in terms of farm.

One last curious thing to mention.  It is an extremely small section of builds, at 5% of the total, but builds that do not take a single point in Dark Pact by level 7 do have a win rate of 54.25% in Very High.  This could easily be just a case of small sample size, but it’s possible that 0/4/2/1 is actually an acceptable build provided you don’t actually need to purge anything vital during the first 7 levels.  I don’t know how confident I’d be in doing that, but it is true that Pounce->Essence Shift does pretty well overall in Very High.  Essence Shift in general seems more important in Very High, which could be a reflection of just smarter aggression and better Pounce usage.

Finally as a bonus, I have the biggest item shifts in Slark by skill bracket.

[Slark]Items

One interesting side note to this is that Vanguard’s win rate drops from 60% in Normal to 50% in Very High.  It’s normal for win rates to drop between Normal and Very High because GPM in Very High is higher in general, but 10% on an item like Vanguard is pretty noteworthy.  In contrast, Vladmir’s Offering only drops from 63% to 61%.  Item win rates never paint the whole story, but what I’m seeing doesn’t give me much confidence in Vanguard Slark, at least in higher skilled games.


The Rise and Fall of Treant Protector

July 6, 2013

 

Published as part of Team Liquid’s Star Ladder S6 Finals Preview

riseandfall

The Rise and Fall of Treant Protector

Yesterday this was going to be a very different article. Then Treant’s 6.78c nerf was announced. It turns out it’s pretty hard to muster the energy to write about competitive Treant when I don’t personally believe that competitive Treant will be a thing for much longer. So let’s instead talk about why Treant became such a dominant force these past two months, why Treant’s nerf will be the largest change of the patch, and how teams might try to compensate to salvage the hero.

Part 1 – The History
There’s this persistent myth that Treant’s rise came out of nowhere, so let me start by dispelling it. Of all the Dota 2 heroes, Treant has arguably changed the most over this past year. In 6.74, Treant nearly had a 60% win rate in public games, and he and Lycan were neck-and-neck for the highest win rate of the patch. Despite this, Treant saw virtually no competitive play. At this point in his history, Living Armor was a passive ability that provided his team armor and HP regen globally during the day and in a 900 radius at night and to buildings. Living Armor as a passive was extremely effective at winning games with virtually no input from the player but only at lower levels of play. This dynamic was problematic, and so Icefrog brought out the hatchet.

In 6.75, Living Armor changed from a passive ability into the global buff we know today, albeit in a much weaker form, and Treant’s win rate dropped an astonishing twenty percentage points to 40%. To be fair, he did receive some other nerfs in this patch. His ultimate Overgrowth lost all of its damage, and he took a hit to his base attack time. Regardless, a twenty point shift in a single patch period is by far the largest I’ve seen in Dota 2’s history.

6.76, 6.76c, and 6.77 all saw further buffs to various aspects of Treant’s stats and abilities, but altogether these buffs only amounted to moving his public win rate to 45%. Finally in 6.77c, Icefrog did the one thing he had been avoiding and buffed the damage block portion of Living Armor.


Living Armor damage reduction is now done for all instances of damage that reduce its charges.


Off this single change alone Treant’s public win rate jumped ten percentage points to 55%. More importantly, unlike his 6.74 incarnation, Treant’s 6.77c win rate rose as you moved to higher skill games. And why not? Using the new Living Armor to its utmost effectiveness required a great degree of map awareness, and if low level pubs were winning quite regularly with it just imagine what an organized team that built their draft around Living Armor could accomplish.

It turns out that we didn’t have to imagine for very long. In May, about a month after the patch, Quantic (then DD) went 5-1 with the hero in the Western International Qualifiers, and from there the hero’s competitive presence skyrocketed. For reference, in April, Treant had a pick or ban rate of 3%. In May, that rate more than tripled to 9.8%. In June it was 52.1%. His win rate over these past two months has been 60.3%. As for Quantic, their record with Treant over these past two months has been 18-4, which comes out to a 81.8% win rate.

Part 2 – The Implications
As I’ve mentioned, Treant’s competitive rise was predicted by his public performances, and looking at those performances in depth we can make two conclusions about the hero.

First, Living Armor defines Treant. I did a breakdown of ~30000 pub games with half taking place before the 6.77c patch and half after. At all skill levels Treant’s public win rates were significantly higher for builds that maxed Living Armor first. While I’d never go so far as to argue that his other three skills are worthless, I would say that there is no hero in Dota 2 right now that is more dependent on a single ability than Treant is on Living Armor.

Second, the most important facet of Living Armor is its damage block. It’s no surprise that Treant’s sudden success was predicated on buffing the mitigation aspect. The life regen and tower healing are nice little perks, but the real strength of the ability lies in the damage block and what it allows your team to do.

Strong Treant comps often revolve around taking a set of strong midgame carries or semi-carries and using Living Armor to turn them into unstoppable monsters during the first fifteen minutes of the match. Two prime examples are Weaver (22-4 when teamed with Treant) and Puck (13-3 when teamed with Treant). Living Armor can be seen as a temporary Vanguard that compensates for the fragility of these heroes while amplifying their evasive survivability.

The biggest mistake then that people make when conceptualizing Living Armor is that they think of it as a defensive ability. Sure it, just like Vanguard, has defensive benefits, but if you’re drafting a Treant (or buying a Vanguard) purely for security then you’re paying too much and would be better served with a different support. The strength of damage block is that it provides an extreme amount of survivability in the early game when hero damage output is still low, so to truly be effective with Treant your team needs to be extremely aggressive to take advantage of the damage block while it is still relevant. With good coordination, Living Armor lets you win trades that you shouldn’t and can turn otherwise suicidal dives into guaranteed kills.

Part 3 – The Nerf
So with that all being said, let’s finally look at the 6.78c nerf to Treant:


Living Armor damage block instances from 7 to 4/5/6/7


Will this be a big deal in pubs? Maybe, but Treant will be fine. He’ll probably see a win rate drop of 3% at the very least, but he has plenty he could stand to lose before becoming an unacceptable pick. Where this nerf really hurts is in competitive games. Why? Because the nerf is harshest during the first two ranks of the ability, and level five often doesn’t come quick for a competitive support.

Let’s take this to numbers. The first level of Living Armor blocks 20 damage per instance. Pre-nerf, that’s 140 damage blocked over the entire course of the spell. Post-nerf, that’s now down to 80 damage. On top of this, you can chew through 4 block instances more quickly than you can chew through 7, so you’ll also be losing some regen. Let’s stipulate that our Living Armor target is taking an attack per second so that he loses 3 seconds of regen post-nerf. In total, Living Armor gave our target 168 HP pre-nerf and 96 HP post-nerf. In effect, the post-nerf version has about 57% of the effectiveness of the pre-nerf version. To cut to the chase, rank two has about 71% effectiveness under these conditions, rank three has 85% effectiveness, and rank four has full effectiveness.

We can take two angles to examine how potentially crippling this is for a Treant comp. For the first, we can look at using a greedy lineup to counter Treant. As an example we have the NaVi vs TongFu game last week where TongFu reacted to the Treant pick by grabbing Enchantress and leaving their Spectre pick somewhat vulnerable with only a Visage support. NaVi responded by devoting too much time towards farming up their Gyrocopter-Rubick-Treant trilane and lost because Enchantress and Spectre dominated the midgame while Living Armor was a total non-factor.

Successful Treant comps are insanely dependent upon midgame aggression. In competitive games in the last two months, Treant is 63-32 (66.32%) in games of thirty minutes or less and 45-40 (52.94%) in games longer than thirty minutes. If you have a mid that’s resistant to dives and draft to get as much as you can out your safe lane and jungle you can very easily out-scale a Treant lineup. Pre-patch, Treant comps could respond to this by getting aggressive and punishing your exposed carry and jungler. With the new Living Armor this aggression is much riskier. On top of that, supports get less farm out of an aggressive trilane, so you risk being stuck with the severely nerfed ranks of Living Armor for that much longer.

The other side to this story is that Treant is now hugely vulnerable to aggressive trilanes. His big contribution to a defensive trilane is, unsurprisingly, Living Armor, and losing nearly 50% of the effectiveness of your primary contribution during those first few levels is going to hurt. On top of that, the aggressive trilane can work to disrupt his pulling and force him to be stuck with lower ranks of Living Armor for a longer period. The longer this goes on, the more Treant is rapidly losing his window of opportunity to influence the game.

Can Treant be salvaged by giving him a solo lane? I have my doubts. In terms of both solo capability and scaling he’s largely inferior to, say, Beastmaster, who is already a niche pickup. He might be able to work as a suicide solo similar to how Tidehunter was played last year, but the current meta doesn’t particularly favor that sort of lineup. Realistically, he might still have legs as the dedicated farming support in a hyperaggressive midgame lineup that seeks to punish a greedy, farm-oriented lineup, but he’ll have a much tighter window of opportunity and his lineups will be less flexible and more vulnerable to targeted bans.

Part 4 – The Conclusion
As of this writing, I don’t know when Valve will release the next patch. If it’s not in effect this weekend, StarLadder might be Treant’s last hurrah (when he’s not banned). If it does go through this week, we might still see some Treant selections, but their success will be dependent on how quickly his opponents adapt to exploiting his new early game weakness. It might not be immediate, but over time the mediocre Treant comps will definitely die off, and Treant will eventually be relegated to a situational pickup at best.

And just as a disclaimer, don’t mistake any of this as complaining . Frankly, I’m surprised the nerf didn’t come a patch earlier. I do think it might end up being excessive and that maybe a nerf to Living Armor combined with some buffs to his other abilities would have been a better way to tone him down while also making him slightly more well rounded. On the other hand, Icefrog probably didn’t want Treant to become the Morphling of TI3, so in that context a harsh nerf is completely understandable.


Part 5 – Bonus Content

Since writing this, I’ve started to second guess myself.  It’s a thing that’s bound to happen when the premise of an article changes drastically just a few days before the promised due date.  I still believe that Treant will eventually fall off, but I’m less convinced that it will be as automatic as I thought it was.

One nerf I had compared this to was Jakiro’s back in late 2012 where he lost 75 damage off Ice Path.  But the more I thought about it the more I came to realize that this was misleading in one important way.  Jakiro is simply going to lose 75 damage off every Ice Path he hits from levels 1 to 3/4.  Nothing about how he plays or how the opponent plays will change.  He’s still going to hit Ice Paths, and those Ice Paths are going to do less damage.

Treant on the other hand is losing potential damage mitigation.  This is only a big deal if the opponent makes it a big deal.  Suppose the enemy mid is constantly receiving Living Armor.  You can try to trade with him, which was definitely a bad idea before the patch, or you can simply try to disengage to the safety of your tower.  Living Armor is less effective now, but it might still be effective enough to make trading a bad idea.  So nerf or not, Living Armor is still an effective deterrent in your mid matchup.

Treant lineups will suffer once opponents are capable of creating lane matchups where level 1 and 2 Living Armor is no longer effective enough, and for the most part this means aggressive trilanes.  If you assume that post-patch Treant is a non-entity and do nothing to hinder his draft or laning, then he’s still quite capable of ruining the game for you.  That’s what I mean by Treants fall not being automatic.

And given that, it’s probably premature to say something like “I don’t personally believe that competitive Treant will be a thing for much longer.”  He’ll still be around for a while; it’s just now he has a huge weakness that teams can exploit.  And of course it’ll take time for teams to create lineups for punishing Treant picks, and even more time for practicing those lineups until they’re comfortable with them.  Until then, he’s still a threat (and should probably see a first round ban for any team playing Quantic).

It also might still be worth looking into specific counters for Treant teams.  I’ve mentioned Bloodseeker before, but with Bloodrage receiving a purge effect last patch he might be a niche counter.  Razor is another interesting option.  He’s seeing more play lately, and he might work as a mid against a lineup that’s trying to push the Treant comp late.  I say this because Unstable Current should proc a purge effect on any aggression and neutralize Treant’s ability to turn mid trades in his favor.  Finally, there’s Rubick.  Treant doesn’t have the strongest options for covering his Living Armor casts, making Spell Steal a potentially equalizing factor in the matchup.  In the past two months, Treant has gone .525 (31-28) against Rubick, which is a pretty decent fall from his .603 win rate overall.