Eternium
Eternium

Announcement

Collapse
No announcement yet.

ANB Bronze #6: Crafting exploits detection

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

    ANB Bronze #6: Crafting exploits detection

    We upgraded our cheating detection algorithms to take into consideration player's crafting statistics. We have enough data to detect exploits.
    Let's put it this way: scientists are able to discover exo-planets light years away by analyzing the spectrum of the star movement. Let's say the challenge of identifying someone exploiting the game by analyzing some simple statistics is orders of magnitude easier. Bonus points, the statistical method doesn't really care about the exploit used, it just flags exploiters and it will flag them for future, undiscovered, exploits as well.

    For sake of transparency, here's a short presentation of the detection algorithm.
    • We simulate millions of crafts assuming highest crafting level and reforges on the weakest stat
    • As a result of the simulation we obtain an average item value distribution for top N items after M crafts. The distribution is normal, having the standard bell shape. The more crafts were made, the narrower the bell (smaller variation/dispersion)
    • Using this distribution we can determine how far from the expected value a player is. And more importantly, what's the likelihood to be that far from the mean
    Rank Hero Class Avg. Item Value Deviation Likelihood
    374 Bounty Hunter 1072.0 17.36 0.0000%
    11 Bounty Hunter 1034.7 16.37 0.0000%
    773 Warrior 1023.5 10.61 0.0000%
    179 Warrior 1007.3 8.82 0.0000%
    6 Mage 1047.7 6.29 0.0000%
    16 Bounty Hunter 1036.1 5.44 0.0000%
    7 Mage 1042.6 4.73 0.0001%
    10 Mage 1046.7 3.91 0.0047%
    3 Bounty Hunter 1045.5 3.81 0.0069%
    4 Mage 1053.9 3.53 0.0208%
    15 Warrior 1053.9 3.53 0.0211%
    7 Warrior 1060.0 3.44 0.0287%
    43 Mage 1055.2 2.96 0.1532%
    6 Warrior 1055.1 2.94 0.1622%
    4 Warrior 1037.8 2.78 0.2730%
    12 Warrior 1042.8 1.91 2.8076%
    6 Bounty Hunter 1015.5 1.59 5.5879%
    12 Bounty Hunter 997.8 1.52 6.3824%
    21 Warrior 1045.7 1.29 9.8631%
    2 Bounty Hunter 1017.8 1.17 12.0881%
    48 Warrior 1011.1 0.98 16.2862%
    32 Warrior 1032.9 0.90 18.3207%
    2 Warrior 1052.7 0.85 19.6844%
    3 Warrior 1036.0 0.75 22.6915%
    Actions:
    • A standard deviation greater than 3 -> Rewards forfeited.
    • A standard deviation greater than 4 -> Permanent ban. If a player is in this situation and thinks that they have wrongfully banned, we strongly recommend them to use their incredible luck by playing their lottery of choice.
    While a 3-sigma deviation is possible, we erred on the side of eliminating false positives, and as such our simulation used an extremely efficient crafter. And also used some other signals to increase our confidence.

    Simply said, if you've made it to this list, consider it as a warning. In the future we will be a lot stricter when detecting exploits. Don't use this numbers as exploiting limits, more signals and statistics are added and an exploiter might find themselves on the wrong side of statistics.

    More importantly, please play the game as it was intended to be played. The crafting exploits used are mainly ruining your game experience. Most of the people on this list have crafted a lot of items, without any exploit you would still have amazing items on your hero. Less tediousness, more fun, in my opinion.

    #2
    Thanks for sharing this, really interesting!

    3-sigma means that there is a 0.27% probability of the result being due to a statistical fluctuation, i.e. great luck and not due to cheating. Given that around 3300 warriors participated in ANB Bronze #6 you'd expect around 9 players to have that kind of luck. I assume that that's where the "other signals" you alluded to come in?
    KUSA GAJE GASU 4903

    CL 2100
    Anisaz, Extra Gold BH, TL 7 (too lazy)
    Keril, confused Mage, TL 7 (can't be bothered)
    Ohad, shieldmaiden, TL 115

    Comment


      #3
      Originally posted by Fish knife View Post
      Thanks for sharing this, really interesting!

      3-sigma means that there is a 0.27% probability of the result being due to a statistical fluctuation, i.e. great luck and not due to cheating. Given that around 3300 warriors participated in ANB Bronze #6 you'd expect around 9 players to have that kind of luck. I assume that that's where the "other signals" you alluded to come in?
      Yes. Other signals and (not shown here) the deviation from the mean in the actual population.

      Comment


        #4
        Thanks for the info...

        Lol some of them with a 10+ deviation. Makes it pretty obvious..

        Good job getting this to work... I might go up a couple places on the leaderboards..

        Comment


          #5
          - I like the approach, using millions of simulations "with highest crafting level", I assume it is done with Mastery Level 50...
          - It is strange to me that the distribution is normal, because of the "Finesse" perk.
          - I guess you need to simulate reforges because the original stat is lost?
          - Also, what about the celestial items? Are the original stats lost as well? If yes, the celestials are simply ignored?
          - Is my guess about how the item value is computed, correct?
          - What about the innate stats (armor, weapon range, block rating)?
          - Are sockets considered, or again, is a bought socket indistinguishable from a crafted socket?

          Anyway, thanks for sharing, and for putting up this detection algorithm. Also, for tuning it away from false positives.


          Mobile Only (iPhone 5 / iPad Air 1st gen / iPad Pro 11" 2nd gen)
          Eternium Files

          Comment


            #6
            There are 10,000+ players in the Bronze event, I am looking forward to whether Silver is 10000+
            Two years ago this number was 40,000+

            Comment

            Working...
            X