Silencer Skill Build Analysis

With Silencer being enabled in both Captain’s Mode and Tournament lobbies, I thought it’d be a good time to examine his popular skill builds.  He also saw a partial overhaul in 6.76, so it’s certainly worth a look at how the public have adjusted their builds in the intervening months.

Before we get to that, I need to mention that I’ve changed my build grouping scheme some.  The big problem I was running into with the Single/Double/Split classifications is that a lot of interesting stuff was happening precisely on the edges of my categories.  Take a 4/4/1/1 build.  It’s fairly common among many heroes, but in my old scheme it could be grouped as a Single Q, Single W, or Double QW depending on the path it took to get there.  There was no easy way of saying “This is classified as a Single Q build, but actually looks a lot like a Double QW build,” and that was a clearly an undesirable outcome.

Instead I’m now using a Primary->Secondary classification.  Primary is what skill gets maxed first by level 8.  Secondary is what skill gets maxed second during levels 8 through 10.  If no skill gets maxed during those periods, I classify it as a split build.  Basically we’re left with 4 primary groups (Q,W,E,Split) and each of the non-Split primary groups having 3 subgroups depending on their secondary (skill 1, skill 2, split).

I re-did the Crystal Maiden analysis using the new method and here’s what it looks like.


The major divisions are bolded and in a larger font.  On the usage chart the sub-categories are measured by the percentage of the primary they represent.  For example, Crystal Maiden Aura(E) builds in Very High use Frostbite(W) as their secondary nearly 40% of the time, so (.20*.38 = .076) 7.6% of the Very High sample is a E->W build.

One useful thing that this build style allows for is that you can easily compare similar Primary/Secondary combinations.  For example W->E and E->W perform very similarly in Very High, with 63.19% and 63.20% win rates respectively.  It also let’s us point out that while Crystal Nova(Q) primary builds as a whole don’t do especially well in Very High, they do seem to do alright so long as they put significant points into E.

On a related note, many of you complained about excessive use of Q/W/E notation in the Crystal Maiden article.  I chose to use Q/W/E in my code so that it could be generalized across all heroes, and since I end up spending a decent amount of time looking at that code thinking of the abilities as Q, W, and E becomes a bit habit forming.

In light of that, I’m going to make a concerted effort to use the names more often in upcoming articles.  However, due to space concerns the tables with still use Q/W/E formatting, and I’ll probably tend to refer to builds in a Primary->Secondary notation using Q/W/E.  Hopefully I’ll actually use the names often enough that these exceptions won’t be a problem.

But enough about that.  Time for Silence.

First a quick ability rundown:

Q – Curse of the Silent: AoE Damage and Mana Drain DoT; Broken by Spellcasting

Scaling — Damage, Mana Drain, Cooldown, and Mana Cost

W – Glaives of Wisdom: Pure Damage Orb Effect; Scales with Intelligence; 1 Point Allows You to Steal Int on Kills and Assists

Scaling — Damage Scaling

E – Last Word: Single Target Nuke with Silence/Disarm Components

Scaling — Damage, CC Duration, Cooldown

His recent patch changes are pretty substantial, so if you’d like to check them out or just want more detailed ability descriptions check the Dota 2 Wiki Silencer Page.

Now to get back to the tables.


Unlike Crystal Maiden, Silencer is all over the place.  Nevertheless, two clear trends stand out immediately.

First, Curse of the Silent(Q) gets both less popular and less effective as you move up the brackets.  Q Primary builds as a whole see the clearest decline in win rate outside of the mostly ignorable Split group.

Second, Last Word(E) builds get way more popular as you move up the brackets.  E Primary goes from just under 1/3 of Silencer’s builds in Normal to nearly 2/3 in Very High.  Q->E and W->E builds also increase in popularity in the upper brackets.

What I think is happening here is that Curse of the Silent(Q) shines in Normal laning.  Partly this is because low level laning has a lot of drawn out 2v2 sidelane battles, which is a much stronger environment for Curse than any lanes involving solos.

On top of that, higher skill players know what to do to counteract Curse of the Silent.  This means that if you want to use it effectively you need to have Last Word(E) there to keep them from just breaking it immediately.  If you have a maxed Last Word and can hit them with Curse just as it triggers, you’re guaranteed a near full-duration Curse, and fully maxed out that Last Word -> Curse combo is worth 690 damage pre-mitigation.  Of course it’s often not going to be that simple, but having a strong Last Word appears to be important in landing long-duration Curses against opponents who know what they’re doing.

In light of this I think we have two build styles.  The first is nuke based and maxes Curse of the Silent(Q) and Last Word(E) in some order.  Curse first seems to work best in lower level games.  Higher level games tend to prioritize Last Word.  One point in Glaives of Wisdom(W) is typically grabbed relatively early to allow for orb-walking and intelligence steal.

The second style is more focused around being a right click semi-carry.  This style maxes Glaives of Wisdom(W) early and usually ignores Curse of the Silent(Q) in favor of some combination of Last Word(E) and Stats.  What’s interesting is that Glaives first gets significantly more successful in the upper brackets.  The most obvious explanation for this is that the the Normal bracket tends to have more carries per team and nuke-based Silencer is a better choice in an environment with low or unreliable farm.    Glaive builds are also likely much more gold dependent and lower level Silencer’s might not have the last hitting to finance an effective Glaive build.  The best performing right click build in Normal appears to be E->W, where you’re still investing into strong lane control and can get a feel for the pace of the game before committing fully into skilling Glaives.

The farm explanation is still speculative of course, but I do have some backing evidence.  I created a measurement that was essentially Silencer’s Level divided by the Average Level of the rest of the team.  Basically if the number was higher Silencer was likely being used in a Semi-Carry role.  If the number was lower, Silencer was being used in a Supportive role.  I then creates the tables based on both the top 40% and bottom 40% to get an idea of how things shifted depending on Silencer’s farm priority.  You can check out the results here, but the summary is that Glaive builds are definitely more common in the Semi-Carry sample.

Like last week we have a by-Point table, but this one isn’t nearly as useful as CM’s


Only two interesting things that stick out to me.

The first is that 1 point in Glaives by 8 is the plurality standard.  The second is that while the trends are mostly worthless, W(Glaives of Wisdom) in Very High is pretty interesting.  In that bracket is appears that you should either max it by 8 or leave it at 1.  Anything in-between underperforms.

Finally we have the clusters, and again, they’re a bit weak.


1-1-3 stands out as the safest overall build, but Curse heavy builds keep up or outperform it in the lower brackets.  Openings that invest heavily in Glaives are enough of a minority that no Glaive-maxing level 5 cluster was popular enough to break the 5% threshold.

Finally, stat builds appear rare, but decently successful.  Most of these are probably Glaives/Stats builds that ignore Curse entirely, but I didn’t peer into this to make sure.  Ult Skipping is relatively rare, but not especially unsuccessful.  Maybe Global Silence is situational enough in public matchmaking that you can afford to hold off on it a level or two if you have other more immediate priorities.  As for Silencer’s overall win rate, it appears to decay a bit, dropping from ~52.5% in Normal and High down to 51.5% in Very High.  This is consistent with the general trend that lane bullying is a much more effective strategy in low level games, and Silencer’s kit is nothing if not annoying in lane.  I don’t think this is in any way an indictment of Silencer’s competitive viability though.  Global Silence is an ult that shines in environments where you’re communicating with your team, and is therefore likely undervalued in public matchmaking.  Besides, it’s not like his Very High win rate is bad, it’s just slightly below his overall win rate.

So to summarize the build philosophy:

If you’re playing in a low level game or don’t have a reliable source of farm, Curse/Last Word builds with a point in Glaives appear to be the strongest.  The priority between Curse and Last Word is situational, but in general Curse performs better in low level games and Last Word performs better in high level games.

Glaive builds are viable if you have a safe lane to farm and have decent last hitting.  Most successful Glaive builds neglect Curse in favor of Last Word and Stats.  But even if you do have good farm, a Curse/Last Word build might be more appropriate for your team comp, so keep that in mind before committing one way or the other.

If you’re not sure what kind of a game you’re looking at, the safest build is to go 1-1-3 and adjust from there.

Edit: I forgot to mention that the sample size for this one was ~5,000 games per bracket.

5 Responses to Silencer Skill Build Analysis

  1. Simo says:

    love your work

  2. Ewig Custos says:

    Awesome. Keep up the good work.

  3. Breezy Idiot says:

    What stats are tracked through the API? Im wondering because I recall reading an article about the development of Halo where the developer tracked player movements and could create a ‘heat map’ of where they spent the most time during testing and used it to do things like add more visual cues to make it more clear which direction to travel in next. If player movement could be tracked and then compared to winrates across different heroes of skill brackets it could be really cool.

    • phantasmal says:

      The API itself is pretty barebones. To be fair, it has to be in order to keep the file sizes reasonable. The API is essentially what Valve and Dotabuff use to create the stats page for each match. It contains a little extra information, like Tower status and now ability orders, and will likely include more things in the near future as Valve adds them. But it’s not going to be the place to go for movement data.

      What you can do is use a replay parser like Bruno’s and go through a lot of the demo files that you can download from Valve. It’s likely that the demo contains most if not all the movement information you’d be looking for, although last I checked Bruno’s development hadn’t gotten that far. It’s a much more elaborate process than API diving, but theoretically you could create lane matchups, gold and exp charts by lane, behavior patterns (roaming, stacking, pulling, etc.) and maybe some kind of positional volatility index you could then use to estimate the point in the game where laning has broken down for each team. Repeat the same tests over enough matches and you could theoretically do even more advanced stuff like determining the strongest average laners by denied XPM/GPM/harass damage/kills or the win rates of Visage in solo/duo/trilanes.

      That’s not to say we’re there yet, but I believe some approximation is possible given the time and resources.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: