Why do you have sanctions/PEP/address checks at the second stage, before manual review — why not after? What technology stack are you familiar with? What do you use at TSB? Where does the case_id come from for the very first events? Can you elaborate on the 3 case sources — what are the second and third ones beyond onboarding? What happens if a doc_verified event arrives before kyc_started, or never arrives? What freshness/completeness SLAs would you set for: ops monitoring, FinCrime dashboard, regulatory reporting? What are the trade-offs between having data near-real-time vs next day? What are your top 3 critical data quality checks for KYC onboarding? Who owns event definitions? Who owns DBT models? Who owns marts? How can you make things compatible with producers when data engineering doesn't own producers? How many engineers do you have, what are you responsible for at TSB? What's the difference between HAVING and WHERE? How do you safely compute a division when the denominator might be zero? What's the difference between INNER JOIN and LEFT JOIN? How do you get the latest row per user? Why is ORDER BY updated_at not enough? How would you investigate the number of incorrectly marked PEP/sanctions checks (false positives) that were later found incorrect by manual review? Which tables, how do you join? How would you get the ratio of onboarding sessions per KYC checks — where do you look in the model? If an attribute needs to be updated over time (e.g., agent reviews happening multiple times), how does that affect your model? What if a customer appeals a rejection and gets re-reviewed by a different agent with a different outcome — how is that reflected? Is it covered? If there are multiple reviews and only the latest is valid, what needs to change?

memorry allocation in k8s pods