Legal-Clearway-Simplification-Dataset

About Dataset

This dataset pairs dense, formal legal/bureaucratic text with plain-language explanations, designed to test and train models on a specific failure mode: simplification that drops legally material details (deadlines, exemptions, penalty amounts) for the sake of readability. Of the 12 rows, 7 are real provisions paraphrased directly from the primary text of India's Right to Information Act, 2005 (cross-checked against the Act as hosted by the Central Information Commission and a Delhi Government legal department mirror) — covering proactive disclosure deadlines (Section 4), the right to not state a reason for a request (Section 6), the 20-year disclosure rule (Section 8), penalty provisions for unreasonable delay (Section 20), appeal deadlines (Section 19), and good-faith legal protection for officials (Section 21). The remaining 5 rows are synthetic, illustrative examples used to test the pipeline's structure and are clearly marked as such. Each row includes the dense original text, a plain-language explanation, a list of key facts the simplification must preserve to be considered faithful, and a provision_type label (exemption, appeal_deadline, penalty, fee_exemption, validity_period, response_deadline, proactive_disclosure, applicant_rights, good_faith_protection). RTI responses and legal notices are legally binding but written in registers most citizens can't parse, and the people most likely to need them are often least likely to have someone available to translate the legal language. This dataset supports scalable simplification while explicitly measuring whether legally material details survive the simplification process — the same place manual simplification by RTI-support NGOs also tends to go wrong under time pressure