From Zero to Open Question: A Controlled Stress Test of Human–AI Scientific Reasoning Under Maximum Transparency Constraints
Abstract (Index version)
This document establishes the methodological and epistemic foundation of the Oxygen Octave Project and the Coherence-Assisted Thinking (CAT) framework.
It documents a controlled experiment exploring whether a non-expert, working under strict transparency and falsifiability constraints, can use general-purpose AI systems to construct and refine a coherent scientific question.
Oxygen was selected as a case study due to its exceptionally well-characterized behavior across chemistry, physics, and biology, allowing structural inconsistencies and hallucinations to be readily detected.
Rather than presenting a scientific hypothesis, this work defines the constraints, reasoning loops, failure modes, and documentation standards of an openly auditable AI-assisted research process.
The document serves as the methodological origin of the project and frames failure itself as a scientifically valuable outcome when fully documented in real time.
