sidhasadhak/beautycommerceos

BeautyCommerceOS是一个大规模合成企业数据仓库，专为在真实业务环境中运行的自主AI代理和分析系统提供基准测试。它模拟了一个现代全球美妆电商公司的完整生命周期，涵盖从用户行为到财务对账的所有环节，并包含了实际企业中常见的模糊性、不一致性和跨职能复杂性。数据集采用现代奖章仓库架构，分为青铜层（原始行为数据，如会话和点击流事件）、白银层（统一维度，如产品和供应商）和黄金层（业务与财务真相，如订单和利润损失）。它旨在评估AI系统在跨领域推理、营销归因、供应链分析、财务对账和实验分析等方面的能力，并故意引入不一致的归因信号、延迟的财务对账等现实复杂性，以增强对推理系统的鲁棒性评估。数据集以Parquet格式存储，总大小小于10 GB，适用于自主代理评估、LLM推理基准测试、分析工程实践等用途，但不适用于真实世界财务预测或生产决策。所有数据均为合成，不包含真实用户或个人信息。BeautyCommerceOS is a large-scale synthetic enterprise data warehouse designed to benchmark autonomous AI agents and analytics systems operating in realistic business environments. It simulates the full lifecycle of a modern global beauty ecommerce company—from user behavior to financial reconciliation—including the ambiguity, inconsistency, and cross-functional complexity found in real enterprises. The dataset follows a modern medallion warehouse architecture with Bronze Layer (raw behavioral data like sessions and clickstream events), Silver Layer (conformed dimensions like products and suppliers), and Gold Layer (business and financial truth like orders and profit & loss). It is designed to evaluate whether AI systems can reason across domains, including marketing attribution, supply chain analytics, financial reconciliation, and experimentation analysis, while intentionally incorporating inconsistent attribution signals, delayed financial reconciliation, and other realistic complexities to reflect real enterprise environments. The data is stored in Parquet format, totals less than 10 GB, and is intended for autonomous agent evaluation, LLM reasoning benchmarks, analytics engineering practice, and similar uses, but not for real-world financial forecasting or production decision-making. All data is fully synthetic with no real user information included.