About the new method of evaluating tank performance

顽童ck强行插入 –ごめんね(´・ω・`)   首先插播一个B站的论平衡性的视频
［各种躺枪就对了］玩游戏被虐的真实原因人艰不拆啊
&或见游戏业产品经理GOUKI-完美的平衡游戏不存在!   f119a君请继续┏ (゜ω゜)=☞
Hello everyone,
this post will be a bit complicated, so bear with me. Basically, a couple of Russian players came up with a new method of measuring whether a tank is overpowered or not (or even underpowered) than classic winrate with some sort of threshold. In the beginning, there was the post by Makkad. This player (LJ) referred to the recent Storm statement that T57 Heavy dropped significantly in winrate and thus doesn’t need to be nerfed. Makkad points out that it practically took a year of being OP to drop to reasonable levels:

大家好，这个帖子会比较长，太长不看请右上角请大家耐心一点。简而言之呢，过去确定一辆坦克是不是OP，用的是胜率加几个参数的做法。几个俄服玩家想到了一个新的手段来测量坦克是不是很OP。由于Storm最近说T57胜率掉的很厉害不必再砍，Makkad发布了这样一个帖子（译注：俄语页面，不懂俄语的不必翻开看）。从中可以看出，要经过一年之久，才能看得出一辆坦克是否真的OP。

A year to discover whether a tank is OP or not based on balance alone is of course too long a period. What is needed is an indicator, that would show the tank OP factor directly after two or three weeeks. In the following part of the post, Makkad proves the fact that such an indicator can indeed exist.

但是花一年时间决定一辆车是不是该砍实在是太长了。所以我们需要找一个方案，两三周就能看出一辆车是不是真的OP。Makkad随后在他的帖子里给出了一个方法。

Based on this premise, Makkad makes another (possibly obviousy) statement, that the tank winrate depends on the ratio of good and bad players playing it. The quality of players playing that respective tank from its introduction shifts gradually: in the first week after its introduction, only extremely dedicated players will have the tier 10 tank, because they buy it for accumulated free XP. Here is the amount of T57′s played in time: in first week it’s only like 340 players with them, while by now there are over 130 thousand of T57′s on RU server

在此基础上，他又提出另外一个观点：一辆坦克的总体胜率取决于大神和坑货玩这辆车的比例如何。从一辆新车出现到全面普及，玩家平均水平会逐渐下降：第一周的时候，只有非常投入的玩家和大挂逼才能立刻买到10级车，因为他们是用换的全局经验砸出来的。下图是T57玩家的数量变化：2012年底T57刚刚在俄服上线的时候，第一周大概只有340个玩家有这辆车；现在？已经有13万辆T57了。

Obviously the first ones to unlock the vehicle are very good players from top clans, but over time, as the vehicle becomes more popular, its winrate spread (when comparing it with players) will correspond the general skill spread of the playerbase, which looks like this for RU server:

显然第一周就能出车的都是大公会的顶级玩家，但是时间长了，随着这辆车越来越多，胜率分布就会逐渐平均下来，和玩家水平的分布图类似，比如下面这样。

Thus it can be concluded that the ratio between good/average/bad players (furthermore referred to as GAB) does influence the vehicle winrate. That’s why the author suggests to use the corellation between the player winrate group and their actual winrate of the vehicle. For example, you take all the players with 52 percent winrate and check their average winrate on the vehicle you are trying to assess. You do that for several of the groups and the result looks like this:

所以好/中/烂玩家水平的比例会影响一辆车的胜率。因此Makkad提出看一看玩家胜率和单车胜率的相关度如何。比如你把所有的52%胜率的玩家都拖出来，然后看看他们玩你想要评判的那辆车胜率如何。比较若干组数据之后，你就会得到大概这样一张图：

The horizontal axis displays the player group winrate, the vertical one the average winrate of that respective group on the assessed vehicle. The dotted line represents the average player winrate growth, while the solid line represents the vehicle. In this case, the vehicle would need a buff, because its winrate grows slower than the winrate of the player groups. This method allows to estimate much earlier, whether the vehicle needs a buff/nerf or not.

横轴是玩家胜率，竖轴是单车胜率。虚线代表假设玩家玩这辆车的胜率应该和他的全局胜率一样，实线则表示实际他们开这辆车的时候胜率如何。上图这辆车就需要buff，因为它的胜率提升速度跟不上玩家水平的提升。

On the picture above, the situation is not so simple. For bad players, the vehicle is hard to play and for good players, the vehicle is overpowered. In this case, the method could even help the developers pick, what to do with the tank. In order to make it easier for bad players to play it, you need to use straightforward buffs (such as armor or alpha increase) and to compensate it by limiting the maximum tank potential (which good players use), for example by nerfing DPM, camo and viewrange (those are the values a bad player will gain little benefit from).

上图则表明不是那么简单。对于坑爹货来说，这辆车很难玩好；对于高玩来说，这辆车非常强力。这样就可以帮助开发者决定要怎么改动车辆的参数——增加装甲、单发伤害之类的参数，就能让坑爹货更容易上手；而削弱那些对高玩有帮助的参数，像DPM，伪装系数和视野，就能限制它在高玩手里的发挥。

To summarize: by looking at not only at winrate, but also at winrate/player quality corellation, you can potentially reach more accurate results. In the comments to this post, Storm stated that he actually passed the method for evaluation to his statistic specialists, as he considers it to be a reasonable idea.

总结一下，就是看一辆车在不同水平的玩家手里发挥程度如何，就能得出更准确的结论。在那个帖子的回复里，Storm说他觉得这个想法不错，已经把这个转发给了他的数据分析砖专家。

The method is applied by another LJ user (Ivanerr) in his post as follows (based on public data from wot-news I assume):

同在LiveJournal的另外一名用户Ivanerr使用这个方案作了他的分析 （注：继续俄文），结果如下

T57 Heavy

Not imbalanced in general – noobs can’t play it, but it is somewhat OP in unicum hands. Ivanerr states that the point, where the two lines cross can be called the “OP point” and the lower it is, the more OP the vehicle is. Ivanerr also advises that the fact the OP point is 52,5 for T57 means that anyone with less winrate should avoid this vehicle, as that person will suck with it.

T57：不算太不平衡，坑货玩不好，但在高玩手里的确有点OP。Ivanerr把两条线的交点叫做“OP点”，两条线相交越早，这辆车越不平衡。他还说，52.5%的OP点表明，全局胜率不到52.5%的人最好不要玩这辆车，因为很可能玩不好。

AMX40

The shape of the T57 graph (“noobs can’t, unicums can pwn”) also means one more thing: that this vehicle can potentially greatly influence the battle. AMX40 seems to be the other extreme case:

AMX40：T57的图表明这辆车影响战局的潜力很大，而AMX40则是另外一个极端：

The fact that noobs pwn with it but unicums don’t means one thing: this vehicle can do very little to influence the battle outcome. It’s a noob friendly vehicle statistically, but unicums will have a hard time influencing the battle with it.

这辆车坑爹货会觉得很不错，而高玩正好相反，说明这辆车对战局的影响力极其有限。统计表明这车对坑爹货比较友善，但是到高玩手里玩不出什么花样。

Foch

Using this method it shows that the Foch is a classic case of an overpowered tank. It has very low OP point, which means even noobs can use it to produce better results than what they usually have.

福煦：福煦是一辆OP的车，OP点很低，说明大多数坑爹货都能发挥出比个人水平更好的效果。（注：不是福三炮）

Maus

Maus is an example of a tank, that is very lackluster. As you can see from the graph, it doesn’t really have an OP point, no matter how skilled you are.

鼠爷是烂车的代表之一。如你所见，这车不管你水平多高，都没有一个OP点可言。

WT E-100

While current winrate results according to developers do not warrant a nerf, using this method the vehicle is completely overpowered

E100运输车：虽然WG根据当前胜率说不能保证会动刀，但是从下图可以看出这辆车明显地OP了。

M48 Patton

Using this method, Patton seems like a noob friendly vehicle, but it does require VERY high skills to truly pwn.

大巴顿：大巴顿看上去对坑爹货还不错（译注：我感觉SS写错了……），但你需要很高的水平才能碾压其它人

KV-1S

Using this method, KV-1S is shown to be completely overpowered. It has a very low OP point, which means even total noobs can pwn with it.

KV1S太OP了。OP点很低，说明坑爹货都能开出去碾压别人。

Rheinmetall Borsig WT

Very, very imbalanced vehicle, it is almost completely overpowered.

莱茵金属搬运车：非常、非常不平衡的车，严重OP了。（译注：请围观战力系统坦克排行→ http://woteff.garphy.com/getTankTop）

T110E3

Slightly tiny little bit OP

110E3：有些OP

T95

Nearly ideally balanced

T95：我们的游戏很平衡

T110E4

Not OP, needs a skilled player

110E4：不算OP，比较吃技术

Jpz E-100

Lackluster, just like the Maus

白兔：烂车一台，就像老鼠

Object 140

A bit underpowered, needs a skilled player

140工程：有一点点差，比较吃技术

FV4202

Mostly underpowered except for the best players using it

FV4202：除了在顶尖玩家的手中都不OP

E100

Pretty fine, not imbalanced

E100：相当不错

IS-4

Quite underpowered

IS-4：相当之差

FV215b (heavy)

Pretty fine, not imbalanced

FV215B：还不错

FV215b (TD)

Again, not overpowered

FV215B 183：不OP

T110E5

Not overpowered, you need to be a skilled player to use it

110E5：不算OP，比较吃技术

AMX50B

A little bit underpowered

50B：有一点点差

Batchat

Has big potential for skilled players, but in noob hands it’s bad. Not overpowered.

查涤纶：在高玩手中潜力无限，在坑爹货手里只能肉侦。总体来说不算OP。

Object 263

Pretty ok, not overpowered

263：挺平衡的

E50M

Very well balanced, neither OP nor UP

E50M：平衡相当好

113 and 121 (Chinese)

Both almost ideally balanced

113和121：平衡看上去相当好

And – IS-7, that is COMPLETELY underpowered

IS-7？烂车一台，太明显了

Regarding the IS-7, Ivanerr has a theory about the influence of players unlocking IS-7 first while they are still noobs, but overall, the stats really don’t look so good.

对于-7这么差的数据，Ivanerr有个想法是因为很多人都把-7作为自己第一辆十级车；但是不管怎么说，这数据看上去都实在太差了

Well, that’s it. What do you think?

你怎么看？