Algorithmic Reparation: Measuring the 'Discursive Gap' Between Retributive and Transitional Justice in Colombia Through Peace Engineering

In the aftermath of internal armed conflicts, the battle for memory is often fought within the semantic structures of judicial archives. This paper introduces the framework of "Computational Forensic Hermeneutics" to analyze the epistemic rupture between the Ordinary Justice system and the Special Jurisdiction for Peace (JEP) in Colombia, specifically regarding the state crime phenomenon known as "False Positives" (extrajudicial executions). Drawing on Nancy Fraser’s theory of Justice as Recognition and Galtung’s concept of Cultural Violence, we argue that the impunity of the conflict was sustained not only by silence but by a "bureaucratic engineering of language" that framed victims as lawful combat casualties. To empirically measure this phenomenon, this research employs Peace Engineering methodologies, utilizing a quasi-experimental design that contrasts three corpus datasets: (1) 280 sentences from the Ordinary Justice system (2002–2008), (2) written Autos from the JEP, and (3) oral transcriptions from JEP recognition hearings. Using Memory-Augmented Neural Networks (Titans architecture) and domain-specific Large Language Models (ConfliBERT), we engineered a "Triangulated Truth Memory" grounded in the narratives of victims (MAFAPO) and Inter-American Court of Human Rights standards. The model calculates the "Discursive Injustice Score" (surprisal metric/cosine distance) for each judicial system. Our findings reveal a statistically significant "Discursive Gap." While ordinary sentences exhibit high semantic distance from the victims' truth—characterized by technical euphemisms and passive voice—the transitional justice mechanisms, particularly oral hearings, demonstrate a convergence toward restorative semantics. This paper posits that the future of global justice in conflict zones relies on integrating Artificial Intelligence not merely for procedural efficiency, but as an auditor of Discursive Justice. We demonstrate how AI can serve as a mechanism for "Algorithmic Reparation," detecting symbolic violence in state archives and validating the epistemic transition from a grammar of war to a semantics of peace