MOBIUSI INCThis benchmark focuses on exception conditions, applicability limits, and negated statements in professional exam rules. Each record provides a question, reference answer, source context, supporting evidence, and evidence location. It highlights details that can be overlooked when general rules dominate retrieval, making it useful for evaluating whether retrieval-augmented generation systems can find the relevant provisions and answer accurately based on the source.
| Field | Type | Description |
|---|---|---|
| file_name | string | File name |
| question | string | The professional exam question to be answered. |
| gold_answer | string | The correct answer grounded in the source context. |
| rule_feature | string | The primary rule detail tested by the question. |
| evidence_quote | string | The exact source wording that supports the answer, including any relevant exception, limitation, or negation. |
| source_context | string | The source rule or passage needed to answer the question. |
| evidence_location | string | A locator for the supporting evidence, such as a document title, article, section, page, or paragraph. |
| Authorization Type | Proprietary - Commercial AI Training License (No Redistribution) |
| Commercial Use | Requires exclusive subscription or authorization contract (monthly or per-invocation charging) |
| Privacy and Anonymization | No PII, no real company names, simulated scenarios follow industry standards |
| Compliance System | Compliant with China's Data Security Law / EU GDPR / supports enterprise data access logs |

Post a request and let data providers reach out to you.
@dataset{Mobiusi2026,
title={Professional Exam Exception and Negation Rules QA Benchmark},
author={MOBIUSI INC},
year={2026},
url={https://www.mobiusi.com/datasets/26d1d4b8de08ae5801b6d66c80aa8364},
urldate={2026-10-22},
keywords={professional exam QA benchmark,exam rule exceptions,negation rule questions,professional exam dataset,RAG evaluation data,open-book QA benchmark,evidence localization,rule detail retrieval},
version={1.0}
}Using this in research? Please cite us.