Scaling National Cybersecurity Assessment with LLM-Assisted Coding: A Validated Comparative Documentary Baseline of Four Andean Countries (2020–2026)

Abstract

National cybersecurity assessment through documentary coding faces a scalability bottleneck: manually coding institutional documents across countries consumes weeks of expert effort, confining comparative studies to single nations. This paper makes a dual contribution. Methodologically, it validates whether a Large Language Model (LLM) can reproduce a structured documentary coding rubric—the Evidence Density Index of the DAM–NC methodology—enabling scalable comparative assessment. Substantively, it applies the validated approach to produce the first comparative cybersecurity documentary baseline of four Andean countries (Ecuador, Colombia, Peru, Chile) from a corpus of 55 verified institutional documents. The validation is three-layered: LLM coding was compared against an independent rule-based reference, replicated across three independent LLMs, and anchored against a human gold standard built by two independent expert coders with excellent inter-coder reliability. The LLM tracked expert human judgment more closely than the rule-based reference, and all substantive regional conclusions proved robust across coding methods and models. The baseline reveals a consistent Govern-to-Recover documentary gradient (Kendall’s W=0.893) and a regional gap in documented recovery capability that survives multi-label coding: no Andean country records post-incident recovery as an institutionalized national instrument, either as a primary or a secondary document function. We interpret this as a documentary gap—a potential indicator of, but not equivalent to, weak operational recovery capability. LLM-assisted coding, embedded in a human-in-the-loop calibration protocol, is a viable accelerator for comparative cybersecurity assessment.

The authors' abstract, as published at the source. Computers, 2026 · DOI ↗