Visual Thinking as a Cognitive Skill: A Pedagogical Framework Beyond Tool Proficiency

The Document Engineering Volume 5 (Diagrams) ships against a substrate that mutates every twelve to twentyfour months: Mermaid releases, D2 layout-engine swaps, TikZ package churn, and from 2024 onward the cheap availability of AI-generated images. A textbook anchored on tool proficiency ages with the substrate; a textbook anchored on visual thinking as a cognitive skill does not. This paper synthesises the visual-thinking, multimedia-learning, dual-coding, diagrammatic-reasoning, design-sketching, and cognitive-load literatures (1966–2026) into a pedagogical framework that recasts visual thinking as a transferable cognitive skill at the level of reasoning and representation, not at the level of stroke economy or rendering fidelity. The framework distinguishes two construct criteria: a cognitive-skill criterion (the construct survives a tool swap, an audience swap, and a five-year horizon) and a toolskill criterion (the construct expires with its substrate). Only the former is admitted as a framework move. Six cognitive moves are specified — Decomposition-to-units, Abstraction-byelision, Relation-mapping, Hierarchy-articulation, Provisionalsketch-then-commit, and Symbol-load-management — each presented with definition, cognitive substrate from at least two independent literatures, observable behaviour at the 14+ learner band, audit indicator at chapter scale, characteristic failure mode, and a worked school-project example anchored in the Vietnamese 14+ context. A four-band progression-by-stratum table maps each move to S0–S3 mastery indicators, supplying the audit loop's input criteria. The framework is engineered as the conceptual anchor for Volume 5 chapters that survive any plausible tool-substrate turnover through 2031, and is paired with audit indicators consumable by the program-level Gemini-asjudge audit loop. Three open questions are flagged: accessibility for non-visual learners, cross-cultural transferability into the Vietnamese learner context, and the AI-image-generation effect on the symbol-load-management move.