Given feature summaries for two card transactions (e.g., merchant category, amount, velocity, geolocation, device fingerprint, account tenure, historical approval/chargeback rates), decide approve or decline for each and justify the decision. Then outline a simple risk strategy for cold-start: propose concrete rules (velocity checks, geo-velocity, MCC limits, block/allow lists, step-up authentication), thresholds, and escalation paths, discussing the loss–friction trade-off and expected impact on approval rate, chargeback rate, and false positive rate. Finally, model adversary behavior: enumerate likely fraudster tactics (card testing, mule addresses, BIN attacks, device spoofing) and explain how you would detect adaptation and update rules or models.