ExplainableMRP via Trace Visibility: Faithful Natural-Language Explanations for Deterministic Planning

ACL ARR 2026 January Submission4763 Authors

Keywords: Deterministic Planning, Trace-Based Explainability, Faithful Natural-Language Generation

Abstract: Material Requirements Planning (MRP) is the cornerstone of manufacturing, yet it often operates as an opaque black box. When shortages occur, planners are left manually sifting through thousands of transactional rows to diagnose the root cause, as current systems provide numerical outcomes without the "why." While Large Language Models (LLMs) promise to bridge this gap, applying LLMs to this deterministic domain risks misinterpretation and hallucination. We argue that reliable explainability in MRP cannot be deduced post hoc. It must be traced. We introduce ExplainableMRP, a framework that transforms the LLM from an independent reasoner into a faithful narrator of the execution trace. By combining schema normalization, explicit trace exposure, and evidence-constrained generation, we demonstrate that our approach resolves input ambiguity and suppresses hallucination to negligible levels. Code will be released upon publication.

Paper Type: Long

Research Area: Interpretability and Analysis of Models for NLP

Research Area Keywords: explanation faithfulness, free-text/natural language explanations, human-subject application-grounded evaluations

Contribution Types: Model analysis & interpretability, NLP engineering experiment

Languages Studied: Korean

Submission Number: 4763