Non-consensus is a core driver of scientific innovation, and its cyclical transformation with consensus constitutes the internal dynamic for scientific progress. As a key carrier for such innovative outcomes, non-consensus papers require more attention and research from the academic community. This paper conducts a systematic exploration of non-consensus papers from the dimensions of conceptual connotation, formation mechanisms, and evolutionary processes. The study finds that non-consensus papers are characterized by high originality, contentiousness, risk, and potential returns. Building on this, the study proposes a suite of feature computation methods, including peer review divergence statistics and text similarity calculation, and discusses their potential applications.