us-tax-law-qa

# US Tax Law Q&A Dataset A synthetic dataset of U.S. federal tax law questions and answers with IRC citation grounding, designed for fine-tuning language models on tax reasoning tasks. ## Dataset Structure | Split | Examples | |-------|----------| | train | 3,500 | | test | 500 | ## Fields | Field | Type | Description | |-------|------|-------------| | `id` | string | Unique example identifier | | `category` | string | Tax law category (international, estate_gift, business_entity, individual, procedure, specialized) | | `subcategory` | string | Specific topic (e.g., controlled_foreign_corporations, gilti) | | `difficulty` | string | basic, intermediate, advanced, or expert | | `question` | string | Tax law question | | `reasoning_trace` | string | Step-by-step reasoning in IRAC format | | `answer` | string | Detailed answer with citations | | `citations` | list[string] | IRC sections cited (e.g., "IRC § 951") | | `is_trap_question` | bool | Whether the question contains a common misconception | | `common_errors` | list[string] | Typical mistakes for this topic | | `edge_case_flags` | list[string] | Notable edge cases or considerations | ## Statistics - **Total Examples:** 4,000 - **Unique Questions:** 98.4% - **With Reasoning Traces:** 100% - **With Citations:** 100% - **Trap Questions:** 11.6% - **Difficulty Distribution:** - Basic: 20.0% - Intermediate: 33.8% - Advanced: 31.1% - Expert: 15.0% - **Category Distribution:** - International: 40.0% - Business Entity: 20.0% - Individual: 15.0% - Estate & Gift: 10.0% - Tax Procedure: 10.0% - Specialized: 5.0% ## Categories - **International tax**: CFCs, Subpart F, GILTI, FDII, foreign tax credits - **Estate and gift tax**: Transfer taxes, valuation, exemptions - **Business entity taxation**: Corporate tax, pass-throughs, S corps - **Individual taxation**: Income, deductions, credits - **Tax procedure**: Filing, penalties, statutes of limitation - **Specialized**: Exempt organizations, retirement plans ## Usage ```python from datasets import load_dataset dataset = load_dataset("DJLougen/us-tax-law-qa") ``` ## Models Trained on This Dataset - [DJLougen/granite4-tax](https://huggingface.co/DJLougen/granite4-tax) - IBM Granite 4.0 Tiny fine-tuned on this data - [DJLougen/qwen3-4b-tax](https://huggingface.co/DJLougen/qwen3-4b-tax) - Qwen3 4B fine-tuned on this data ## Limitations - Synthetically generated; may contain errors in legal reasoning or citations - Not a substitute for professional tax advice - Focused on U.S. federal tax law only - Template-based generation; question diversity is bounded by template count per subcategory