Privacy reproducibility package

Full reproducibility package for the AMCIS 2026 paper "Benchmarking Privacy Compliance in EU Crypto-Asset Markets: A Quantitative Analysis of GDPR and MiCA Implementation" by Cordoba Otalora, Khalid, and Themistocleous (University of Nicosia; University of Agder / Singular Logic). The study is the first systematic quantitative benchmark of Crypto-Asset Service Provider (CASP) privacy compliance in the early post-MiCA period. Privacy policies of 24 CASPs across 12 EU member states were retrieved between 3 and 16 March 2025, coded against a 14-item rubric (5 categories: Transparency & Clarity, Consent & User Rights, Crypto-Specific Measures, Data Security & Transfers, Accountability) with category weights 20/25/30/15/10 percent, and analysed using composite scoring, cross-country ANOVA, repeated-measures ANOVA, and Pearson correlations against regulatory-maturity indicators. The initial sampling pool contained 27 CASPs; the analytic sample was reduced to 24 by capping at most four CASPs per country to prevent any single jurisdiction from dominating the analysis. This package is the end-to-end trail from raw policy text to final composite score. It contains (a) the 14-item scoring instrument with regulatory anchors and category weights (01_rubric.csv); (b) 24 structured analysis files, one per CASP, with score, one-line analytical rationale, and verbatim justification quote for each of the 14 rubric items (analysis/entity*a.json); (c) the flattened 24 × 14 = 336-row item-level scoring matrix (07_item_level_matrix.csv); (d) 24 archived privacy-policy texts used as coding inputs (documents/); (e) entity-level composite scores with ranks (02_entity_scores.csv); (f) category and country aggregates, correlations with regulatory/market variables, and four-scheme weighting-sensitivity analysis (03–06); (g) the three figures published in the paper; and (h) full analysis notes for reproduction in Python or R. Coding protocol: an LLM-assisted first-pass coder produced an initial score, analytical rationale, and verbatim justification quote for each item; two human coders then independently reviewed and revised every score against the policy text. Inter-coder reliability on a 20 percent double-coded subsample reached Cohen's kappa ≥ 0.80 across all items (Landis and Koch 1977), with consensus resolution of discrepancies. All 24 composite scores reported in the paper reproduce exactly (to ±0.01) from the item-level matrix supplied here. The package is released under CC-BY-4.0.本完整复现包对应将于AMCIS 2026发表的论文《欧盟加密资产市场隐私合规性基准测试：通用数据保护条例（General Data Protection Regulation, GDPR）与加密资产市场监管条例（Markets in Crypto-Assets Regulation, MiCA）实施情况定量分析》，作者为Cordoba Otalora、Khalid与Themistocleous（尼科西亚大学；阿格德尔大学/奇异逻辑公司）。 本研究是首个针对MiCA实施后早期阶段加密资产服务提供商（Crypto-Asset Service Provider, CASP）隐私合规性的系统性定量基准研究。研究团队于2025年3月3日至16日期间，收集了欧盟12个成员国共24家加密资产服务提供商的隐私政策，并依据14项评分框架（涵盖5大类：透明度与明晰性、同意与用户权利、加密资产专属措施、数据安全与传输、问责机制）进行编码，各类别权重分别为20%、25%、30%、15%、10%；随后采用综合评分法、跨国家方差分析（ANOVA）、重复测量方差分析以及与监管成熟度指标的皮尔逊相关性分析开展研究。初始采样池包含27家加密资产服务提供商，为避免单一司法辖区主导分析，设置每个国家最多纳入4家提供商的限制，最终分析样本缩减至24家。 本复现包涵盖从原始政策文本到最终综合评分的全流程溯源材料，具体包括：(a) 含监管锚点与类别权重的14项评分工具文件（01_rubric.csv）；(b) 24份结构化分析文件，对应每家加密资产服务提供商，每份文件包含14项评分框架条目对应的得分、单句分析理由与逐字引用的政策文本佐证（analysis/entity*a.json）；(c) 扁平化的24×14共336行的条目级评分矩阵文件（07_item_level_matrix.csv）；(d) 24份用作编码输入的加密资产服务提供商隐私政策存档文本（documents/文件夹）；(e) 包含排名信息的实体级综合评分文件（02_entity_scores.csv）；(f) 类别与国家层面的聚合数据、与监管/市场变量的相关性分析结果，以及四组权重敏感性分析结果（对应文件03–06）；(g) 论文中发表的3幅图表；(h) 可通过Python或R复现分析的完整研究笔记。 编码流程说明：首先由大语言模型（Large Language Model, LLM）辅助完成首轮编码，为每个条目生成初始得分、分析理由与逐字佐证引用；随后由两名人工编码员独立对照隐私政策文本对所有得分进行审核与修订。在20%的双重编码子样本中，编码员间信度达到科恩κ系数（Cohen's kappa）≥0.80（依据Landis与Koch 1977年的研究标准），并通过协商解决分歧。 论文中报告的全部24项综合评分均可通过本包提供的条目级评分矩阵精确复现（误差范围±0.01）。本复现包采用CC-BY-4.0协议开源发布。