Terminology management is key to quality assurance in translation projects, yet translations often remain stiff and imprecise in Chinese – English translation of official texts, even when termbases and machine intervention are in place. This study proposes a dual-layer (form – concept) consistency model: whereas form-level management unifies the renderings of identical forms, high-frequency formulaic expressions in official texts exhibit form-concept misalignment, so that form-level consistency may itself cause concept-level inconsistency. Three types of misalignment are deduced: referential drift, semantic compression, and collocational misalignment, and are validated through a case study of an English translation project of a COVID-19 situation report. This study further formalizes a terminological concept-relation network and upgrades termbase records from form – rendering pairs to “form-concept-contextual feature-rendering” quadruples, turning terminological restructuring into a reusable mechanism. Semantic relation identification is shown to be a necessary complement to consistency management and provides a knowledge interface for human – machine collaborative terminology management.