Diagnose 4% weekly revenue drop using history

You only have a single table weekly_revenue(week_start_date DATE, revenue_usd NUMERIC) containing the last 104 weeks for Instacart. A DS reports that the most recent week’s revenue dropped 4% vs the prior week; assume no measurement error. Design a rigorous analysis to decide whether this is expected noise or a true anomaly, and quantify it. Specifically: (1) Define your baseline and formal null/alternative hypotheses using only this weekly series. (2) Propose at least two forecasting/uncertainty approaches (e.g., seasonal-naive by week-of-year vs last year’s same week; STL or ETS/BSTS without covariates) and a decision rule (e.g., residual z-score or whether the observed value lies outside a 95% prediction interval). (3) Explain how you will handle seasonality and trend with only weekly aggregates (no external covariates), including weeks with movable holidays. (4) Show how you’d decompose the observed 4% into trend, seasonal, and irregular components and communicate risk if it lies within expected variability. (5) Now suppose you later receive weekly aggregates by geo for orders, average order value (AOV), active users, and marketing spend. Outline a drilldown to attribute drivers (e.g., orders×AOV decomposition, geo-wise contributions, hierarchical shrinkage across geos), and specify how you’d detect whether the drop is localized or systemic.