
Alberto A.⭕️
17.5K posts

Alberto A.⭕️
@culinaryphysics
Aristologist in training & a food blogger @ https://t.co/pNOOzsFGOk. Tweeting abt FOOD, Culinary Physics, guns, blockchain, PenTest, #Axie & #STEPN




Mystic axie draw scheduled for NEXT WEEK! A few weeks ago, we surpassed Community Milestone 2 in Jin’s Fortune Spin Axie Machine. This unlocked a Mystic axie draw — and we’re almost ready to kick it off! Here’s what’s happening 🧵👇







Yesterday, I asked CT to act as a fraud analyst on a challenging case. The results were clear: most thought we should request more info or block the transaction. Let’s go deeper: Quick reminder on the case context: - Experienced crypto user - Multiple flags e.g. recent 2FA fails, logins from 4+ countries - Requested biggest ever withdrawal from user’s account to a fresh wallet What did we do? due to the multiple flags, we put a hold on the transaction and requested more info about the transaction from the user. When we couldn't reach them for 90+ minutes, we declined the transaction and restricted the account to protect the user in case their account was compromised. Plot twist: The transaction was legitimate, and we ended up causing the user frustrating friction. Impact: - User had to go through account recovery process - Had to contact support to get account unlocked - Shared their frustrations on social media as soon as their transaction was held - Had to re-submit their transaction once account was unlocked - The entire process took multiple hours to resolve Our intention here was to protect the user from unauthorized account access and withdrawal. Objectively, this was a 'successful' fraud prevention, where we followed best practices (hold → investigate → block if unresponsive). But in this case, the process led to a result that rightfully frustrated a legitimate user. It’s my team’s job to minimize these hurdles as much as possible, constantly improve, and make the recovery experience way-better. Another challenge is the cost of being wrong goes both ways: - Fail to block fraud = user loses life savings, loses trust in crypto - Restrict a legitimate transaction = user loses trust, potential public backlash Crypto fraud prevention is notoriously hard (and it’s why some exchanges limit crypto sends), and we're always looking for ways to iterate and improve our protections. The reality is: - While perfect accuracy doesn't exist, we want to get much better - Every decision has trade-offs - Context is everything, but often missing with crypto sends compared to other payment rails - We have a varied and wide user base and need to expand from one size fits all models and approaches - Social escalations can give us extra perspectives on what users may experience I joined @coinbase to help balance these difficult trade-offs and to keep delivering the best possible experience to all of our customers. We’ve already decreased account restrictions by >80% and in the last few weeks we've made significant improvements in the fraud models responsible for such cases (~20% improvement). We've also made infrastructure improvements that allow us to continuously retrain several of our models, accelerating the speed at which we can iterate and improve. How do you think we could have handled this better? What would good friction look like to you? Also, if you find these challenges interesting, come and work on them! We’re hiring, @CoinbasePltfrm

















