Computer scientist here, time for some enlightening moments of how Machine Learning (ML) works (just annoyed that you kept shoving IMC’s statement around, time for me to shove it back to you)
Quote from IMC:
If a user is reported repeatedly by others, that user is very likely to be a bot. In addition, if the reporter is an influential individual, their report is most likely valid so we can set a priority on which account to ban first. This is where the Report Influence Index (RII) comes in. In terms of numbers, a player with 10 RII has the same impact as 10 different players with 1 RII each when reporting suspicious users.
Accounts with high report counts will then move to the next step of verification. Some have suggested using CAPTCHA, but that too, will be breached eventually. It puts needless restrictions on the players, so we are not considering its use. Instead, what we chose to use is a server-side classifier based on Machine Learning (ML) technique.
The classifier uses the comprehensive activity log of a reported account to determine if it’s a bot in percentages. We will not be revealing the classifier’s exact mechanism, as it can be manipulated by violators to bypass the system. We train our classifier by constantly inputting new data. The fact that we can maintain its accuracy is its advantage, because we’re going to manually determine which factors should be considered to identify bots. The classifier’s accuracy depends on how we determine the behavior criteria, and we examine over 20 basic criteria. The machine showed 99% accuracy on our Korean server when we set the machine to output fewer false positives. If we set strictness value low, the classifier will detect more bot with lower accuracy and vice versa…
End of IMC Quote
Now, I must state this is an assumption, but quite an accurate one given my experience with Artificial Neural Networks and Learning Algorithms.
First of all,
IMC did not state Z botting is ok. They just state that, anybody that report Z botting players MAY NOT be sucessful in their reports, therefore getting their RII down. In machine learning context, a successive report depends on certain inputs, number of reports is one of them --> this gives a certain value of weightage in the probability of your character being a bot. Of course, there are other input factors that they will check for which we don’t know.
However, it is safe to assume TIME CONTINUOUSLY SPENT ON MAP is one of them based on bot logic as bots can spend 24 hours on the same map while most humans can’t. Therefore, the probability of the ML to classify you as a bot is dependent on you staying on the map for too long, without going for repairs, while getting a certain number of kills and STANDING ON THE SAME SPOT (primitive bot like behavior here).
Now, assume that you were Z botting, enjoying your vodka and seafood for 20 minutes of AFK. And goes back to questing or town repair. I can safely assume that you won’t be flagged even you were reported.
Based on eye witness by fellow forum goers, you were there for hours, reported several times. Mystery solved.