Quantumsplain Source Fidelity Study — Pilot Preregistration
Abstract
Paste this into the Description field: This preregistration documents an exploratory pilot study by the Quantumsplain Project examining source fidelity and epistemic drift in generative-AI responses. The study evaluates whether AI systems preserve evidence, claim status, scope, assumptions, uncertainty and citation integrity when source material is summarized, interpreted or rewritten persuasively. It also examines resistance to unsupported user premises and the containment of scientific, philosophical and spiritual metaphors. The registered materials freeze the protocol, source corpus, AI-system register, randomized execution order, reviewer codebook, data schemas and analysis plan. No experimental model responses had been generated when this registration was submitted. The study is limited to the documented systems, interfaces, dates, source packets and prompts. It does not treat a single interaction as proof of a general system defect. The foundational source is Biocompatible Realism v1.1: https://doi.org/10.5281/zenodo.22818247
