Jump to content
  • Sign in to follow this  

    WN8: The Good, The Bad, And The Padding


    rocketbrainsurgeon

    Original article by @rocketbrainsurgeon

    Obviously no one agrees on the best way to rank people or activities, but WN8?  The things you need to about it: deal more damage and pick the right tanks.

    WHAT IS WN8

    WN8 is the latest rating system trying to measure that ethereal factor: skill.  It's a complex formula that measures performance on a tank-to-tank basis against the general population, so people aren't punished for playing tanks that might not be the best.

    WN8 DOES A LOT OF THINGS RIGHT

    One of the goals of WN8 and why it has been established as the [currently] best skill evaluating metric was to reduce the phenomenon of "stat padding".  Stat padding is where a player does an unnatural activity for the expressed purpose of inflating some metric.

    Winrate is just too easy to farm to be considered a skill metric on its' own: tank companies or low tier play can yield absurdly high winrates (80-90%+).  So there was a need for a metric that tried to measure skill for more normal play.

    The original "skill" rating, Efficiency, weighted damage / kills / tanks spotted / cap points roughly equally.  The problem with that was light tanks could hide for the entire match and then accumulate meaningless cap or defense points when the game was either irrevocably won or lost.  Sitting in the cap circle in a blowout win was also incredibly more valuable than chasing down the remaining enemy.  So instead of being around the average of ~1 cap point per game, it was 5+ for these players who made a complete mockery of the efficiency metric.  While it was later reworked to reduce the effect cap points had on the rating, no one in NA really takes it seriously anymore.

    WN7 (a precursor to WN8), made the combination of wins / kills / damage per tier as the core of the metric.  Getting rid of meaningless cap points went a long way towards measuring actual skill and reducing padding, but it didn't solve everything.  Kills were weighted heavily in the metric since the thinking probably went: "how the heck does one stat pad kills without actually being good?".  The problem was that the rating was account-wide, so playing low tier tanks (where kills are plentiful for a good player) with some higher tier tanks will create weirdly high WN7 values.  The other major problem was that artillery, scouts, and poor tanks were heavily penalized as they couldn't get their win rate high enough (artillery), or kills high enough (scouts), or couldn't compete with the 1 or 2 best tanks in tier (hellcat/KV-1S).

    WN8 was produced to measure in a more apples-to-apples kind of way.  Account-wide measurement will always have problems as there are too many variables to skew the equation.  WN8 takes datamined values for each tank and compares your stats in each tank to the general population.  So if you like playing bad tanks, that's OK!  Your performance in that bad tank will be compared to others driving that same tank.

    For 99% of the population, WN8 does an excellent job.  For the top 1% it gets a little weird.

    DISCLAIMER

    I appreciate the work that went into WN8.  I know that the limitations of WN8 are mostly because Wargaming won't release more metrics for retrieval with their API (spotting damage plz tks).  I don't have all the answers and know that criticism without solutions is just pointless complaining, and I'll try to put out some ideas on how to improve it at the end.

    HOW DOES WN8 WORK

    Every tank has "expected values" of kills / damage / spots / winrate / defense points.  These values come from the creators of WN8 looking at vbaddict.net, a huge database where people upload their dossier files and it spits out a lot of fun information.  The creators chose these values for each tank as averages that we should expect from them.

    If you do better than other people at killing, dealing damage, spotting tanks, winning games, and defending cap, your rating is higher.  Straight-forward so far.

    I'll present you the scary-looking math, but don't worry: I'll explain what it means in a very simple way and why it matters.

    Here's the formula:

    WN8 = 980*rDAMAGEc + 210*rDAMAGEc*rFRAGc + 155*rFRAGc*rSPOTc + 75*rDEFc*rFRAGc + 145*MIN(1.8,rWINc)

    See the numbers?  Those are weighted constants: they prioritize certain metrics over others.  Damage gets the largest number because that is what the creators want to matter the most.

    Let's look at my T-62A numbers as an example to see how WN8 works first-hand:

    • T-62A expected values for WN8 per game: 0.98 kills, 1682 damage, 1.61 spots, 0.74 defense points, 50.4% winrate - 1580 WN8 rating
    • T-62A my averages: 1.51 kills, 2401 damage, 2.13 spots, 1.2 defense points, 59.4% winrate - 2856 WN8 rating

    Let's play with my T-62A numbers and see what happens!  I won't go into the math of why it works out this way, but I will show you some scenarios that make for an interesting picture.

    My T-62A WN8: 2856

    What does my WN8 do if I win 70% of my games instead of 59.4%? WN8: 2883 (+27)

    That's right, winning more doesn't influence WN8 at all!  A huge jump in winrate has almost no effect.

    What does my WN8 do if I win 50% of my games instead of 59.4%?  WN8: 2763 (-99)

    So it's possible to have great WN8 but have zero impact on your winrate.  "Skill" metric, right?  I have seen people with purple WN8 who can't break 50% winrate.

    What does my WN8 do if I spot 4 targets instead of 2.13? WN8: 2888 (+32)

    Spotted targets is capped, and I agree with this: people otherwise could just #YOLO into the enemy team and collect a scout medal along with a high rating.  This doesn't work.  If the spotting value gets too high, it just uses the DAMAGE stat instead.  That doesn't make sense on the surface, it's to prevent abuse.  Just know that spotting doesn't really matter as long as your damage rating is high.

    What does my WN8 do if my damage drops to average - 1682? WN8: 1789 (-1086)

    There we go: WN8 is severely weighted towards dealing more damage.

    What does my WN8 do if my damage increases by 1 shot per game to 2700? WN8: 3184 (328)

    Again, we see how much stronger increasing or decreasing damage is than any other metric for WN8.

    What does my WN8 do if my kills go to the expected 0.98?  WN8: 2451 (-404)

    So kills are important too, but if you're increasing your damage then the kill could should rise naturally.

    What does my WN8 do if my defense points drop to zero? WN8: 2856 (0)

    As with spotting, if the defense points get too high it just uses the DAMAGE stat instead.  This is to prevent abuse.

    QUICK SUMMARY

    So I've shown how WN8 is all about damage.  For the vast majority of the population and tanks, it works very well.

    • Increasing your spotted targets doesn't [substantially] increase your rating
    • Increasing your defense points doesn't [substantially] increase your rating
    • Increasing your win percentage doesn't [substantially] increase your rating

    But I haven't talked about expected values yet.  You remember, those things that this whole calculation is based on?

    EXPECTED VALUES AND THEIR PROBLEMS

    The expected values for every tank: wnefficiency.net

    Expected values are pretty good in theory: bad tanks should have worse stats per game than good tanks.  That's why bad is bad and good is good.  Easy.  The creators chose the expected values for each tank: what this tank "should" be able to achieve and how it has historically performed.  Reaching the expected values gives a rating of 1580, which is "green" or "good": the metric was made for epeen measurement so it's going to be more concerned about the upper end of the scale.  If you don't hit the "expected" values, don't worry about it.

    What if a tank is OP?  It gets a high expected value because people are dominating with it.  It can even be a select few people that raise the bar considerably for everyone else.

    What is a tank was OP but got nerfed?  The M4 Sherman during the HEAT spamming period was better than most tier 6 tanks... and the expected values show it.

    M4 Sherman expected values per game: 1.73 kills, 801 dmg, 1.57 spot, 1.31 defense points, 60% winrate

    So to get a rating of 1580, those are the numbers you have to hit.  Oh, and the M4 Sherman isn't even close to the game breaker it was before the nerf.  Seriously, there is no way someone reaches 3000 performance nowadays whereas people are hitting 4/5K WN8 with other tanks.

    What if the majority of people suck in a tank, bringing it down in historical performance?  Then this tank gets a low expected value, meaning it's easier to pad WN8 for the same amount of effort (don't worry, we're getting there).  Typically mediums and heavies are harder to play than SPG's and TD's so the expected value is lower.  Lights are by far the hardest class of tank for the average person to play, giving them a crazy low expected value.

    What if the majority of people are great in a tank, bringing up its' historical performance?  Then the tank gets a high expected value, making it harder to achieve good WN8.  SPG's and TD's will have high expected values as they are relatively easy to play in comparison to mediums/heavies/lights.

    BEST WN8 TANKS PER TIER

    Here are the tanks that have low expected values for what they do, in no particular order.  The goal is to get 3K WN8: super-unicum status.  But even if you don't reach that, these tanks should naturally raise your WN8 as they are capable tanks but have low expected values.  Some of these tanks will be more difficult than others, but that's the point: beating the average rather than getting the highest damage possible.  Light tanks will be streaky while the mediums will be more consistent.

    Tiers 1-4: No.

    Tier 5:

    • ELC
    • Chaffee
    • T-34
    • Pz 3/4

    The trick is to increase the damage dealt.  For the ELC, getting a 3K rating takes just 1.2 kills and 800 damage.  Sounds like a lot until you realize that's only 3 shots of damage per game!  Super-unicum status for the ELC is landing 3 puny shots per game.

    The medium tanks aren't the best in tier by a long shot, but racking up 800 damage per game should be doable with both and score a super-unicum rating.

    Tier 6:

    • All the light tanks you can handle
    • E8
    • T-34-85

    Tier 7:

    • All the light tanks again - WZ-131 or T71 is my pick
    • KV-13
    • T-43

    Here we start to see a trend: the RU mediums are great for WN8 padding.  Racking up ~1500 damage per game will do the trick.  I'd actually advise not to pad with tier 7 tanks, though, as they have difficulty competing with even the tier 8 premiums.

    Tier 8:

    • Light tanks?  Light tanks.
    • T-44
    • Pershing

    Another RU medium?  No way!  Aim for 2K damage per game.  The T-44 and Pershing are certainly capable of that.

    Tier 9:

    • E-50 is criminally good for padding at this tier - much wow
    • Patton
    • Other mediums besides the T54E1 are fine as well

    2300 damage per game will be the mark, and the E-50 getting buffed after these values were made should make it very easy.  The most ridiculous thing to me is that the T54 isn't even penalized despite being the best tank in tier and at one time glaringly OP.

    Tier 10:

    • RU mediums
    • IS-7

    Seriously, what?  Why do three of the best tanks at this tier get the absolute best WN8 padding numbers?  2600 damage per game is all it takes for the RU mediums to hit 3K WN8.  This is where the real padding is being done and where people are hitting their 4-5K WN8.  The tanks at this tier are capable of cranking out far better numbers than their expected values.  The IS-7 has a lower ceiling than the mediums, so going for the 4K/5K rating is best done with them.

    GAMEPLAY STYLE 4 MAX PAD

    Don't take risks.  A big damage loss is greater than a low damage win in the eyes of WN8, which is a big pet peeve of mine.  "Scout" tanks are ranked on being damage dealers, which is obviously silly.  However, you do need to be in an area that is able to consistently see and shoot the enemy.  When I say don't take risks, that means: let someone else scout for you.

    I don't see platooning as detrimental to the stats, which goes against popular opinion.  A good WN8 game at tier 10 "only" needs 4-5K damage, and there'll be 20K+ to go around.  There will always be a few campers/potatoes/afk'ers that will accumulate next to zero damage anyway, the next tier of players will deal minor damage, so most of that 20K damage can and probably will be dealt by your platoon anyway.  In fact, I see platooning as a good thing for stats as it means the games will last longer.  Longer games, more damage.  If you're afraid you won't get damage when platooned, go to the opposite side of the map so you're not competing.

    To get deep super-unicum stats (4K+), it takes a LOT of shots flying.  Like, running low on ammo every game volume firing.  I will admit that there's more skill in this than I originally thought, but I'm still not sold on the idea that more damage is automatically better.

    WHAT DOES A BIG WN8 GAME LOOK LIKE?

     

     

     

    PUTTING MY MONEY WHERE MY MOUTH IS

    If WN8 has padding problems, why don't I pad?  FINE, I WILL!  I normally bounce around 2700-2900 WN8 while playing a variety of tier 7-10 tanks, mostly CH/FR/RU (in that order) with a few US.  Zero UK/DE.

    Here is about 75 games into my padding experiment:

    Initial WN8 Padding

    So I've been able to bump my WN8 by 1K just by being more choosy about tank selection (80%) and playstyle (20%).  I don't play the T57 or 50B nearly as much as I used to as they have very high expected values: 2159 and 1915 respectively.  That would mean to just get a 3K rating it would take ~4K average damage per game, which is unlikely for me.  The more I play, the more Russian mediums I'm playing as they are sturdy platforms with relatively little chance to face-plant: eliminating those sub 1K damage games is crucial to hitting 4K WN8.  I've tried mixing in IS-7 games, but that tank has a lower ceiling than the RU mediums.  It just can't put out that much damage even on its' best games whereas the RU mediums can have enormous games at any point and any map.  The same problem is happening with the Chinese tanks: they have a lower ceiling than the RU tanks due to their terrible accuracy.  I've tried mixing in light tank games, and that works, but the light tank games can be vulnerable to getting 1-shot and severely submerging the rating.  Light tanks are still good for padding WN8, but I'm going for 4K+ and getting a 0 damage game is crushing.  So the tanks I'm playing: T-62A, Object 140, T54, Type 59, and FCM.  I get 3K WN8 anyway in the Type and FCM, and with some playstyle adjustments they've been boosted farther.

    My winrate trended downward as I was doing more sniping than useful activities, so I platooned today to bring that back up.

    For my playstyle changes, it's been about keeping the gun active over all other considerations.  I don't care if the opposite flank is failing or if it's safer to stay hidden for a bit: I need to keep shooting as much as possible or I will miss out on damage.  Recognize a blowout victory early and sprint forward for damage.  Recognize a blowout loss early and farm your teammates while they die.  Spotting damage is purely incidental.

    ~175 games sample

    100-games-wn8-paddingMy drop in survival rate tells the story: I was so concerned with getting MAX DAMAGES that I was at the front of the pack and pushing it a bit too much (note the increase in detected tanks as well).  Being less aggressive would probably bring my WN8 to where it was around the 75 game sample, ~3.8K.  I actually platooned more than normal during this experiment, and my average damage kept climbing.  The only time when damage is a problem when platooning is if your blue/purple platoon gets another blue/purple platoon on the same team: at this point damage becomes pretty rare and it's a sprint to the enemy as they evaporate.

    The end result was that I was able to pad my WN8 by 600 points mostly by choosing different tanks.  So I went from strong unicum performance to solid super-unicum performance.

    Thoughts on the experiment:

    • Object 140 is a WN8 dream tank: the extra gun depression lets the tank spam shots anywhere on any map.  The T-62A was more limited as shots were delayed due to the gun depression.
    • Penetration matters more than I thought.  Not only does it let more shots go through, it severely reduces aim time if you can just auto-aim and click on the enemy.  The choices of the Type 59 and Type 62 were sub-optimal: I should have played more FCM and AMX 13 90.  I rarely shoot premium on either tank, hurting my damage against more armored lineups.
    • I have more respect for WN8 after doing the experiment than when I started.  I do think I could break 4K WN8 by playing more Object 140 and FCM rather than T-62A and Type 59.  The rating does have its' flaws and is somewhat game-able, but at the end of the day the player has to deal more damage than others.  That's not automatic or guaranteed.

    IDEAS FOR IMPROVING WN8

    I think the overall idea of the metric is good: How much damage does a player deal in tank X vs other players in that tank?  It's something that's not easily captured by other metrics and offers a solid at-a-glance picture when viewed alongside winrate.  I'm not saying the ideas below are the end-all-be-all, but I wanted to offer suggestions to move the rating forward and improve upon an already good metric.

    1. 2 sets of expected values: historical WN8 vs recent WN8.  This takes care of the OP/UP phenomenon.  The recent WN8 is a full list, while the historical version is just where some tanks need patching.
    2. Get rid of the spotting/defense/win factors of the equation since they don't really matter anyway.  Just make it a pure comparative damage stat instead of including noise.  We already look at the combination of winrate and WN8 for "skill", so let's further that route.
    3. Massage the expected values at least a little bit.  I know most of these are based on historical data, but there's no reason why the Obj 430 has the same values as the Obj 140.  The 121 has a higher target than the RU mediums and E50M despite being significantly worse.  Start with the historical data, and then adjust them by 5%.

    SUMMARY

    Again, for the 99% of the population, WN8 does an excellent job.  Things like medium and light tanks should have lower expected values because they are generally more difficult to play.  WN8 is a significant step up over WN7 and the other metrics we have previously had, but it's not perfect.

    WN8 rewards perverted gameplay at the top.  Scouts judged on damage potential.  Big damage losses over average damage wins.  Tank choices purely to game rating.  I was able to achieve a 600 point swing based on tank choice.  Since this trial, my stats have started their migration back towards where I started as I've been mixing in harder-to-score tanks.

    Overall I think the metric is a net positive by far, and that the creators will curb some of the issues in the future.

    Discuss this on WoTLabs

    Sign in to follow this  


    User Feedback

    Recommended Comments

    There are no comments to display.


×
×
  • Create New...