Missouri statute parsers diverge at 0.43 false-negative rate on numeric thresholds; 93.2% of held-out chapters fail survival certificate under global error model
The paper quantifies extractor disagreement on statutory logic and supplies a statistical certificate that passes only under narrow calibration conditions. It reveals that 93.2% of chapters become unusable when error rates are transferred globally, exposing a core limitation of automated legal reasoning. The work supplies both a deployable test and evidence that current machine-extracted statutes remain fragile for high-stakes use.
Future statutory parsers will require jurisdiction-specific disagreement measurement before any implication is treated as stable input to compliance engines or case-law retrieval systems.
Saka et al.: Fewer than 15% of production statutory extractors will pass the survival certificate on new U.S. state codes without per-chapter recalibration by end of 2027.
Sources (2)
- [1]Primary Source(https://arxiv.org/abs/2609.01741)
- [2]Supporting Source(https://arxiv.org/abs/1609.08144)