Diagnose rising delivery cost precisely

A restaurant sees a notable rise in delivery cost per order over the past month. Delivery cost is narrowly defined as the expense from courier pickup at the restaurant through customer drop-off, covering courier fees, platform fees, and delivery-related refunds or adjustments. It deliberately leaves out food ingredients, kitchen labor, packaging materials, and dine-in costs. You have historical records: orders(order_id, order_ts, customer_id, store_id, distance_km, promised_eta_min, actual_eta_min, platform, city_zone, courier_id, delivered_bool, cancel_reason), courier_payments(courier_id, order_id, pay_amount, surge_multiplier), fees(order_id, platform_fee, refund_amount, promo_amount), weather_by_zone(date, city_zone, precip_mm, temp_c, wind_kph), and store_ops(store_id, open_hours_json, staffing_level, kitchen_prep_time_avg_min). Design a rigorous analysis plan to:

1. Check the metric definition and verify the increase is not caused by scope creep or a change in the denominator.
2. Break down delivery cost per order into drivers (mix: platform, distance bands, zones, time-of-day; rate: pay per km/min, surge; execution: cancellations, reattempts, SLA breaches).
3. Identify causal hypotheses (e.g., weather shocks, staffing changes that raise wait-at-pickup, platform policy changes) and measure each with suitable methods (e.g., difference-in-differences across unaffected zones, fixed-effects regression, or event studies).
4. Suggest at least two actionable experiments to lower cost (e.g., batching, zone remapping, pickup-wait time SLA), including success metrics, guardrails, power calculations, and expected sources of bias.
5. List common pitfalls particular to this case (e.g., accidentally including packaging/food costs, treating tips as costs, double-counting refunded orders, survivorship bias from dropping undelivered orders), and how you will prevent them. Overview: This question tests a data scientist's ability in metric validation, driver decomposition, causal inference, and experimental design for operational delivery cost analysis.