Refresh target: turn a messy transactions table into an RFM-style EDA with three
charts and three stated insights. Follow Submission_assignments/playbooks/EDA.md.
1. Clean first, plot second
Parse dates, coerce numerics, dedupe on the order key, and quarantine (never silently drop) rows with missing customer ids. Reshape with merge/pivot/melt: one row per customer with Recency/Frequency/Monetary columns.
2. The three charts
- Distribution of Monetary per customer (log-scale if skewed) — where is the mass?
- Recency vs Frequency scatter — do recent buyers buy more often?
- Monthly revenue trend with annotation of the largest drop — what changed?
3. Insight discipline
Each chart gets one falsifiable sentence: direction, magnitude, and the segment it applies to. "Revenue fell" is not an insight; "Top-decile customers drove 62% of the March dip after promo end" is.
Dataset: homepage_ab_test.csv for the drill; ecommerce transactions for the mini.
Recap
Clean (parse/coerce/dedupe/quarantine), reshape to one-row-per-customer RFM, plot distribution + recency-frequency + trend, and write one falsifiable sentence per chart.