Foundation Principles for Document Engineering in the AI Era: A Minimum Viable Principle Set Stable for Five Years

The Document Engineering textbook series (Vol 0– 5: Foundation, File Systems, Markdown, LaTeX, Citations, Diagrams), aimed at Vietnamese learners aged 14 and up, requires a principles anchor that survives the half-life of the tools it teaches. This paper synthesises the cognitive-load, desirable-difficulty, multimedia-learning, large-language-model empirical, and academic-writing literatures (2003–2026) into a minimum viable principle set for document engineering pedagogy in the AI era. The set contains ten principles: load-aware sequencing; preserved generative friction; verification-beforedelegation; concrete-then-symbolic representation; orchestration over generation; tool-half-life isolation; citation provenance as first-class data; visual semantics over visual decoration; reflective revision with named audience; and falsifiable craft criteria. Each principle is presented with definition, theoretical basis, auditrubric criteria sufficient for one auditor to score one volume chapter, mapping to a specific applied-volume chapter, and characteristic failure mode. Each principle is engineered to be falsifiable: a chapter either satisfies the audit rubric or it does not, and the failure mode names the specific mechanism by which a chapter fails. The set is argued to remain stable for at least five years on the basis that every principle is anchored in cognitive architecture, learning theory, or epistemic-quality criteria that pre-date the tooling cycle. Three open questions are flagged for downstream Layer H research: how to weight the principles when the audit rubric encounters genuine trade-offs, whether the orchestration principle survives the next agentic-hierarchy generation intact, and whether the Vietnamese learner context introduces a country-specific eleventh principle.