When Can a Machine Trust a Statute? A Survival Certificate for Machine-Extracted Legal Logic

机器何时能信任法律条文？一种针对机器提取法律逻辑的“生存证明”

Abstract: Statutes are increasingly parsed by machines before people read them, and the parsers disagree: on Missouri’s statutes, two independently written extractors diverge on numeric-threshold presence at a false-negative rate of 0.43. We ask what formal logic survives such noise.

摘要： 在人类阅读法律条文之前，机器对其进行解析的情况正日益增多，但不同的解析器之间往往存在分歧：以密苏里州的法律条文为例，两个独立编写的提取器在识别数值阈值时，其假阴性率（false-negative rate）差异高达 0.43。我们探讨了在如此严重的噪声干扰下，什么样的形式逻辑能够幸存。

We build a passive survival certificate for the Duquenne-Guigues implication basis of machine-extracted statutory contexts: per-attribute inter-extractor disagreement is measured, replayed against the basis in 1,000 Monte Carlo trials, and an implication is certified only when a one-sided Wilson 95% lower bound on survival reaches 0.95; every certified implication carries premise spans and a minimal counterexample.

我们为机器提取的法律语境构建了一种针对 Duquenne-Guigues 蕴含基（implication basis）的被动生存证明：通过测量各属性在不同提取器间的分歧，并在 1,000 次蒙特卡洛模拟中针对该基进行重演；只有当生存率的单侧 Wilson 95% 置信下界达到 0.95 时，该蕴含关系才会被认证；每一项被认证的蕴含关系都附带前提范围（premise spans）和一个最小反例。

On 29,365 Missouri sections and 502 Indian central-Act sections, the preregistered held-out gate passes (10 statute families across 7 Titles exact; 16 across 11 with 5% tolerance), yet under one globally deployed error model 93.2% of held-out chapters fall below the informativeness floor, and a 2x2 factorial assigns that to calibration-rate transfer, not selection.

在对 29,365 条密苏里州法律条款和 502 条印度中央法案条款的测试中，预注册的留出集（held-out gate）通过了验证（在 7 个法典标题下精确匹配 10 个法律族群；在 11 个标题下以 5% 的容差匹配 16 个族群）。然而，在一种全球部署的错误模型下，93.2% 的留出章节未能达到信息量阈值，通过 2x2 析因分析表明，这归因于校准率迁移（calibration-rate transfer），而非选择偏差。

The certificate is usable but fragile: deploy it per-chapter-calibrated or error-tolerant. Code, data products, and the audit trail, including one retracted claim, are released.

该证明虽然可用但较为脆弱：建议按章节进行校准部署或采用容错部署。目前，相关代码、数据产品及审计追踪（包括一项已撤回的声明）均已发布。