Problem
Design a top-K popular items shelf that ranks menu items per restaurant based on finalized purchase counts
- Correct late refunds and delays via scoped batch reruns instead of modifying live counts
- Compute fixed windows cheaply by aggregating daily totals instead of scanning raw history on reads
- Serve consistent rankings under high skew by exposing atomic versioned snapshots with freshness metadata
Key Test Points
- Immutable partitioned facts
- Precomputed daily rollups
- Atomic snapshot publication
Spot an issue or have a suggestion?