On-Demand Randomized Question Generation (ORQG): A Decentralized, Cryptographically Secure Framework for Leak-Resistant Examination Systems

The integrity of large-scale academic and competitive examination systems continues to be underminedby question-paper leaks. Conventional examination workflows depend on a pre-compiled paper thatpasses through a long chain of custody — authoring, moderation, printing, physical transport, and strongroom storage — each stage introducing an independent point of exposure. This paper presents OnDemand Randomized Question Generation (ORQG), a decentralized, cryptographically secure assessmentframework in which the final examination instrument is not compiled in advance. Instead, a unique paperis synthesized locally at each examination center, 15–30 minutes before the test begins, using aCryptographically Secure Pseudo-Random Number Generator (CSPRNG) applied to a large, encrypted,continuously growing central question bank under a fixed psychometric blueprint. We formalize thegeneration process with set-theoretic and combinatorial models, specify a CSPRNG-based referenceimplementation using Python's secrets module, and present a formal security analysis mapped tostandard properties (confidentiality, integrity, availability, authentication, auditability, and nonrepudiation) together with an explicit threat model and its limits. We report results from a Monte Carlosimulation of 500 independently generated papers drawn from a 90,000-item pool, measuring generationlatency, inter-paper overlap (Jaccard similarity), and the combinatorial entropy of the design space. Wefurther develop an Item Response Theory (IRT)-based normalization procedure to preserve scorecomparability across non-identical papers, and we discuss the residual risks the framework does noteliminate. Rather than claiming that leakage becomes impossible, we show that ORQG shifts the attacksurface from a single, high-value, long-lived artifact to many short-lived, low-value, isolated artifacts, andwe quantify by how much the associated risk is reduced under stated assumptions.