SyntheticMonthlyFinancials-JP-v1.0.0: 94 fictional firms, 703,036 journal lines

The dataset was created because publicly available financial data is mostly annual or quarterly, making it hard to see monthly cash movements or trace figures back to individual transactions. Real journal-level data is rarely disclosed due to confidentiality. The developer combined LLM-generated company profiles with a Python-based monthly simulation to produce consistent journal entries and statements, avoiding the imbalances that arise when LLMs write numbers directly. The accompanying QA pairs, generated with OpenAI's gpt-6-pro, are designed to test how well AI models can read financial data across multiple files. All 275,318 transactions passed double-entry checks, and balance sheet, income statement, and cash flow statement consistency held for all 94 companies across all 12 months, according to the developer's verification. However, the data is synthetic and simplified—corporate taxes are booked only in March, there is only one year of history, and the answers have not been independently audited.