Change only starter/eda.py. The dataset is scenario/shop-export/transactions.csv — a shop export with exact duplicates, missing/unparseable amounts, and one bad date.

• load_transactions(path) -> pandas.DataFrame: reads the CSV.
• clean_transactions(df) -> (clean_df, stats):
  • Drop exact duplicate rows, count them as duplicates_removed.
  • Parse amount to numeric; unparseable rows are removed and counted as missing_amount.
  • Parse date; unparseable rows are removed and counted as bad_dates.
  • stats = {"duplicates_removed": int, "missing_amount": int, "bad_dates": int}; clean rows keep parsed amount and date.
• summarize(clean_df) -> dict returns:
  • rows_clean, total_revenue (rounded to 2 decimals), revenue_by_region (each value rounded to 2 decimals), top_category (category with the highest total revenue), first_date / last_date (ISO YYYY-MM-DD).
• write_results(results, dest): JSON, UTF-8, indent 2, trailing newline.
• main(argv): runs on the shipped scenario CSV, merges rows_total, duplicates_removed, missing_amount, bad_dates into the results, writes scenario/shop-export/results.json, prints rows_total=... rows_clean=..., returns 0.

Done means python verify.py starter exits 0 with all twelve tests passing, and python verify.py starter --expect-failure exits nonzero because the starter no longer reproduces the unfinished state.