Design and Technical Validation of an Offline-First Configurable Digital Sand Table for Power-Grid Emergency Training

Abstract

Power-grid emergency training requires a system that can encode operating rules, expose the consequences of time-critical decisions, and run in constrained network environments. This paper presents an offline-first configurable digital sand table implemented as a browser-local scenario engine. The architecture separates instructor-authored scenario content from runtime logic through a JSON-based model and executes each exercise as a finite-state process. An explainable assessment pipeline records actions, response latency, timeouts, dimensional impacts, and rule-based feedback. An irreversible error ceiling prevents later gains from restoring a perfect score after an incorrect decision. Technical validation was conducted on an anti-icing emergency prototype through static dependency inspection, schema workflow checks, state-boundary analysis, and scoring-invariant tests. The 190,728-byte core runtime contains no external script, external stylesheet, fetch, XMLHttpRequest, or WebSocket dependency; it therefore executes from a single local HTML file after browser loading. Tests confirmed bounded state updates and the full-score invariant across representative error counts. The result is a lightweight, auditable architecture for configurable emergency exercises rather than a fixed electronic questionnaire. This work contributes a reproducible design pattern for offline-capable training platforms in critical-infrastructure domains where network availability cannot be guaranteed. By decoupling content authoring from execution logic, the proposed framework enables rapid scenario iteration while preserving deterministic assessment outcomes. The finite-state formulation ensures that every learner action maps to a verifiable system transition, supporting post-exercise auditing and regulatory compliance. Future work will extend the model to multi-role collaborative exercises and integrate adaptive difficulty scaling based on learner performance profiles.