M5 Walmart sales streamed through Kafka and Spark Structured Streaming, 23 May to 19 June 2016 · 4 California stores, 12,196 SKU-stores. Stock levels, costs and the e-commerce share are simulated; the demand is real. Web version of the Power BI report.
Units sold ÷ (sold + lost) · week 1 runs on simulated opening stock
All 28 days · selected store highlighted
Sorted by WAPE, worst first · stock-capped = sales limited by stock that day
A challenger is promoted only if it beats the champion on the latest week
STOCKOUT is raised at end of day, STOCKOUT_INTRADAY during the day, DEMAND_ANOMALY when a department is more than 25% off its forecast.