Brief
Publish a notebook EDA on ecommerce transactions following the EDA playbook: cleaned frame, one-row-per-customer RFM table, 3 charts, 3 falsifiable insights.
Dataset
Ecommerce transactions CSV (train visible; holdout customer slice for the insight-generality check).
Starter guardrails (88)
- Start from the EDA-playbook skeleton: clean -> reshape -> plot -> insights.
- Charts must render from a fresh kernel run top-to-bottom (no hidden state).
Deterministic checks
Notebook executes end-to-end; RFM table has one row per customer; 3 chart artifacts present; no hardcoded insight numbers (recompute from data).
100-pt rubric
- Cleaning correctness (parse/coerce/dedupe/quarantine): 25
- RFM reshape validity: 15
- Chart quality (labels, scale choice, annotation): 25
- Insight falsifiability (direction + magnitude + segment): 25
- Fresh-run reproducibility: 10