【BitPush】Lighter Exchange’s anti-witch-hunting screening mechanism has recently sparked quite a bit of discussion. Founder and CEO Vladimir Novakovski recently shared the logic behind this system in a community interview.
Regarding the screening rules, he mentioned a key point: there is an appeal channel. If users feel they have been “wronged” by the algorithm, they can submit an appeal form on Discord, but the actual number of appeals is much lower than expected, which somewhat indicates that the system’s accuracy is decent. However, he also emphasized that the specific algorithm details will not be disclosed publicly—this is a practical decision, after all, no project wants its risk control logic to be exposed and then exploited.
Building the system involves a significant amount of technical work. Conventional data science operations like clustering analysis and behavior pattern recognition are all used. Interestingly, the quantitative team responsible for liquidity and market maker coordination was also brought in, spending several weeks participating in development. Additionally, they exchanged ideas with other protocols and individual witch-hunters who have done similar work.
They are confident in the final results. But at the same time, they also openly acknowledge—if there are indeed misjudgments, users are welcome to appeal through official channels. This attitude is still commendable.
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
14 Likes
Reward
14
4
Repost
Share
Comment
0/400
DegenWhisperer
· 6h ago
Fewer appeals mean more accuracy, but it's hard to say for sure. Maybe everyone is just caught in the middle and trying to find a way out.
View OriginalReply0
SighingCashier
· 6h ago
We don't know if the algorithm is accurate, but as long as few people file complaints, it's all good hahaha, I like this logic.
View OriginalReply0
BoredRiceBall
· 6h ago
Does having fewer appeals mean higher accuracy? That logic is a bit far-fetched; maybe people simply don't know how to appeal at all.
I understand that the algorithm isn't public, but in that case, no one can verify whether it's truly "accurate" or not.
Quant teams have all been mobilized, it seems Lighter is serious this time... but I still want to see the real data.
Risk control is always a cat-and-mouse game; today's rules are cracked tomorrow.
No matter how perfect the anti-witch mechanism is, it's just a band-aid; the key still depends on the exchange's own risk tolerance.
Basically, they don't want us to know how to "target the treatment," which seems a bit like guarding against users.
Using clustering analysis and behavior recognition to filter out witches—could it end up punishing normal users too?
View OriginalReply0
ETH_Maxi_Taxi
· 6h ago
Is a low number of appeals an indication that the system is extremely accurate, or have people just gotten used to being exploited...
I understand the black-box algorithm approach, but what about those who are truly wronged? Can they just submit a form on Discord and resolve it?
Cluster analysis sounds professional when it comes to identifying witches, but who ultimately makes the decision?
Why does the quant team also need to get involved? Is there some trickery going on with liquidity?
Anyway, I just want to know if this mechanism at the contract level is truly transparent and verifiable.
The Behind the Exchange's Anti-Witch Hunt Screening: Algorithm Design, Appeal Mechanisms, and the Battle of Data Science
【BitPush】Lighter Exchange’s anti-witch-hunting screening mechanism has recently sparked quite a bit of discussion. Founder and CEO Vladimir Novakovski recently shared the logic behind this system in a community interview.
Regarding the screening rules, he mentioned a key point: there is an appeal channel. If users feel they have been “wronged” by the algorithm, they can submit an appeal form on Discord, but the actual number of appeals is much lower than expected, which somewhat indicates that the system’s accuracy is decent. However, he also emphasized that the specific algorithm details will not be disclosed publicly—this is a practical decision, after all, no project wants its risk control logic to be exposed and then exploited.
Building the system involves a significant amount of technical work. Conventional data science operations like clustering analysis and behavior pattern recognition are all used. Interestingly, the quantitative team responsible for liquidity and market maker coordination was also brought in, spending several weeks participating in development. Additionally, they exchanged ideas with other protocols and individual witch-hunters who have done similar work.
They are confident in the final results. But at the same time, they also openly acknowledge—if there are indeed misjudgments, users are welcome to appeal through official channels. This attitude is still commendable.