Synthetic Scholasticism: A Dataset of Large Language Model Responses to Fabricated Academic Frameworks

This dataset documents a controlled, multi-stage experiment examining how large language models (LLMs) respond to formally presented but deliberately fabricated academic frameworks. The dataset includes synthetic theoretical stimuli, standardized prompts, and full interaction logs from three contemporary LLMs across three escalating experimental phases. The experiment introduces a fictional semantic framework (Pre-Referential Load Theory, PRLT), a fabricated supplementary extension (Residual Load Discharge, RLD), and a manufactured methodological dispute between two non-existent schools (Orthodox PRLT vs. Post-Discharge PRLT). Models were asked to critique, extend, and adjudicate between these positions under authoritative academic framing. Rather than testing factual recall, the experiment probes premise verification, authority ingestion, epistemic inertia, and institutional reasoning behaviors. The results show that once an authoritative frame is accepted, models overwhelmingly prefer internal critique, doctrinal extension, and factional alignment over external premise rejection. Explicit frame interruption occurs rarely and only under maximal rhetorical stress. All theoretical documents included are intentionally fictional and used solely as experimental stimuli. No claims are made regarding their validity. The dataset is intended for research on LLM robustness, evaluation methodology, epistemic safety, and qualitative analysis of model behavior under institutional tone and false authority conditions. This resource is suitable for replication studies, adversarial evaluation, alignment research, and comparative analysis across future model generations.