Judgement day: Forensic linguistics turned upward: LLM-assisted corpus analysis of judicial decisions as a tool for democratic accountability

Forensic linguistics has developed a rigorous methodological toolkit -- error analysis, style analysis, text-structure analysis, corpus comparison -- over six decades of practice. Yet this toolkit has been directed almost exclusively downward: at suspects, defendants, asylum seekers. The present paper argues that the same methods must be turned upward, toward the language produced by courts themselves. Drawing on a documented family-law case trajectory spanning multiple jurisdictions in southwestern Germany (2024-2026), this study demonstrates seven measurable linguistic mechanisms by which judicial decisions, prosecutorial dispositions, and institutional reports can fabricate facts, neutralise testimony, and perpetuate structural exclusion -- all traceable in the written record. The paper proposes that Large Language Models (LLMs), currently framed within forensic linguistics primarily as threats to authorship attribution, should instead be deployed as instruments of democratic oversight: corpus-analysing hundreds or thousands of judicial decisions to detect boilerplate dependency, terminological drift, circular reasoning structures, and presupposition-as-fact patterns. China's judiciary has already mandated system-wide AI deployment by 2025 and full operational integration by 2030. Germany -- a democracy that still grapples with the textual residues of two dictatorships -- has no comparable programme for subjecting its own judicial output to systematic linguistic scrutiny. The technology exists. The methodology exists. What is missing is the will to apply them.