The Potentiality Dictionary: A Dynamic Provenance-Aware Data Framework for AI Meaning and Action in the Roof–ROFT–Root Architecture

Existing dictionaries and terminology resources provide essential infrastructures for system- atizing established words, concepts, meanings, and usages. However, comparatively fewer systems are designed to continuously capture and validate emerging forms, newly differen- tiated senses of existing forms, insufficiently structured conceptual relationships, and their action-result records arising from creative language, reading, conversation, field observa- tion, AI, and Physical AI experience as reusable AI data. This paper proposes the Potential- ity Dictionary, a dynamic provenance-aware editorial and information-system framework for addressing that gap through the Roof-ROFT-Root architecture. The Potentiality Dictionary does not replace general-purpose dictionaries or neologism lists. It selectively investigates potentiality-oriented Terms, Senses, Concepts, and Relationships whose contribution cannot be reduced to a restatement of an established sense or to a one-off contextual effect. The editorial lifecycle is designed as a cycle of Quick Capture, AI-assisted prior research and recommendation, researcher candidate selection, deep validation, approval, Roof canon- ical registration, Root empirical accumulation, and re-review. Version 1.0 minimizes collector interpretation during Quick Capture and prioritizes the observed term, actual usage, source context, and verifiable source evidence. When a printed book is directly used as the source, photographs of the cover, colophon/copyright page, and the full page containing the usage are linked as evidence. For digital and field sources, in-app direct photography or screen capture is preferred when feasible. Original images remain immutable, while OCR, crops, summaries, and AI-extracted metadata are stored as derivatives. Continuous collection of complete audio or video is not the default because of storage, privacy, and copyright risks; minimum necessary evidence and source identifiers are preferred. Before candidate selection, AI investigates dictionary definitions, specialized terminology, prior usages, and relevant literature to assess whether the captured meaning is sufficiently explained by existing Senses. Evidence, comparison logic, uncertainty, model version, and rule version remain auditable and are stored separately from researcher decisions. The data model separates Term, Sense, and Concept, not as a claim that these three form one canoni- cal standard triad, but as a project-specific integration of concept-designation distinctions in terminology work with lexical sense/concept distinctions in machine-readable lexical models. Definition is initially treated as a representative descriptive attribute of Concept. Contextual novelty alone does not establish a new Sense. A new Sense candidate requires that estab- lished Senses be insufficient to explain the core meaning and that the difference remain in- dependently describable outside the immediate utterance. The system also records whether novelty occurs at the Term, Sense, Concept, or Relationship level, and it keeps novel-sense status separate from potentiality-entry status. Each canonical entry is structured by five research frames: what it is; how it acts; what results it produces; what value arises from those results; and how those results can be eval- uated. The second frame, action, is an umbrella category that may be decomposed into realization, implementation, and execution depending on the object. Within the Potentiality Dictionary system, Roof DB stores validated canonical data and research state, while Root DB stores selectively related reality-state references and empirical AI usage, behavior, and result records. ROFT is not an internal mediator between these databases; it is the con- trolled exchange boundary between the internal dictionary system and external AI/Physical AI. This v1.0 is a design-science research artifact rather than a claim of completed AI per- formance validation. Subsequent work will implement the database and capture application and conduct controlled studies of semantic interpretation, creative language generation, and contextual decision-making.