Capstone · Two rounds · Notebook audit
A retailer needs next month's sales forecast. Check five notebook decisions, then revisit the information boundaries when labels arrive late.
Loading the notebook audit…
Round 1 of 2 · Decision 1 of 5
A retailer needs a monthly sales forecast for each of its 24 existing stores. Each row is one store-month; the target is units sold that month. This fictional notebook has a strong training fit. Check what it was allowed to know before using its October forecast.
Forecasts are issued at 00:00 on the first day of each month. Stores close at 20:00; sales totals and actual promotion spending are finalized at 23:30 on the last day. Promotion plans are locked at 18:00 on the preceding month’s last day, and those original snapshots are retained. All times use the same local time zone.
Development rows cover January 2023 through September 2025. October 2025 is the final test month, with outcomes hidden during development. Replay each historical forecast using what was known at that time. Compare mean absolute error in units sold, weighting every store-month equally; keep this metric fixed.
Commit all five decisions before explanations. You can go back before committing.
Use the archived promotion plan, the previous month’s finalized sales, and the forecast month’s actual promotion spending.
Choices stay on this page until you leave or start a new attempt.