A large model needs fast daily refreshes, but old transactions can be corrected or deleted. How would you keep it accurate?

Instruction: Senior-level, open-ended case. Several approaches can be defensible; explain your assumptions, recommendation, tradeoffs and what evidence would change your choice. Use a correction policy you can explain. State whether history is mutable and whether the environment supports selective partition refresh; do not claim that change detection discovers every old update or deletion. For a mock interview, allow two minutes for your first answer, then 30 seconds to respond to the pressure-test twist. State what changed, your next action and what you cannot safely promise.

Context: This open-ended senior case assesses decision quality. Compares a simpler broader window with targeted historical repair, exposes the deletion and moved-date risks, and makes accuracy and operational cost measurable.

Updated

Official answer available

Read the opening below, then unlock the full answer and practical guidance.

I’d separate the freshness requirement from the correction requirement. We need to know how far back changes can occur, whether transaction dates can move, how deletions are recorded, and how quickly a correction must appear. A successful refresh tells us the job completed; it doesn’t prove that older partitions contain the latest business truth...

Upgrade to view official answer

Related Questions