Persistent misconceptions about scientific concepts develop early in elementary education and resist traditional teaching. The diagnostic tools most commonly used to identify them, such as multiple-choice and tiered tests, reveal that an answer is incorrect but not the reasoning behind it. This article presents DIALOGOS (DIAlogic Learning Observatory for the Grasping Of Science), a conceptual framework for a teacher-guided conversational tool that elicits and maps elementary pupils' science misconceptions through natural dialogue. The framework was developed through a theory synthesis of 68 sources identified in Scopus, ERIC, and Google Scholar, comprising 62 peer-reviewed works, two preprints, three books, and one policy report, managed in Zotero and hand-coded in a five-field extraction matrix. Coded claims were integrated through conceptual translation, cross-strand confrontation, and constraint propagation across three areas: conceptual change theory, dialogic teaching, and educational chatbot research. The synthesis yields six design principles and a four-part architecture: a dialogic elicitation engine grounded in Socratic questioning and the elicit–confront–resolve sequence; a mapping module comparing pupils' reasoning against a research-based catalogue and a concept-map overlay; an individual misconception profile; and a class-level observatory for teachers. The framework repositions educational chatbots from answer providers to diagnostic listeners, shifts misconception diagnosis from item-based to conversation-based mapping, and specifies six testable propositions for empirical validation