A Synthetic Multi-Store Retail Point-of-Sale Transaction Dataset in Square POS Export Format

This dataset contains 970,838 fully synthetic retail point-of-sale (POS) transaction line items, representing a 30-store U.S. retail chain over two calendar years (January 1, 2024 – December 31, 2025). It replicates the structure of Square POS "Items Detail CSV" exports, with one row per item and 22 fields covering transaction details, products, pricing, discounts, taxes, payment methods, store information, customer attribution, refunds, unit cost, and gross profit. A key feature of the dataset is its documented ground-truth retail event calendar. Sales were generated using a layered demand model that incorporates store volume tiers, day-of-week effects, seasonality, month-level trends, 5.5% year-over-year growth, and major U.S. retail events such as Black Friday, Cyber Monday, Super Saturday, Valentine's Day, Mother's/Father's Day, Independence Day, Back-to-School, Christmas closures, and post-Christmas return surges. The accompanying ground_truth_event_calendar.csv records all daily multipliers, discounts, refund rates, and store closures, enabling validation of forecasting and machine learning models. Files: (1) square_item_sales_detail_24mo.csv.gz — main line-item dataset; (2) ground_truth_event_calendar.csv — per-day applied multipliers, discount/refund probabilities, and closures; (3) square_item_sales_summary_by_store_24mo.csv and (4) square_sales_summary_by_store_24mo.csv — aggregated views; (5) validation_report.json — realized statistics; (6) generate_square_dataset_v2.py — the seeded generation script (seed 42) enabling bit-for-bit reproduction; (7) full documentation (data dictionary and generation methodology). Potential uses include time-series forecasting with chronological train/test designs (each calendar event occurs twice), event-effect estimation validated against ground truth, interpretability/feature-attribution benchmarking, store clustering, market-basket analysis, margin and discount analytics, anomaly detection, and teaching retail analytics without privacy constraints. Provenance: All records were generated using a seeded Python simulation with no real customer or business data. The generation process and documentation were developed with AI assistance and reviewed by the author. The dataset reproduces only the publicly documented structure of Square POS exports and is not affiliated with, endorsed by, or derived from Square or Block, Inc.