Before Scientific Emergence Becomes Visible: Independent Evidence for Structural Precursors in Large-Scale Scholarly Knowledge

Can the organization of scientific knowledge carry detectable traces of future emergence before a research area becomes visibly successful? We test this question with PBSK, a temporally locked, reproducible science-of-science framework motivated by the seven-stage arc in The Impossible Idea: How New Realities Enter the World: possibility, question, pattern, imagination, insight, discovery, and transformation. The analysis separates a 12-landmark exploratory cohort from an independent power-validation cohort comprising 120 future-emergent OpenAlex topics and 240 pre-cutoff matched controls. All predictors are restricted to information available by 2018, while outcomes are defined over 2019–2023. In the independent cohort, pre-specified cross-domain collision was higher five years before emergence (paired d = 0.266, permutation p = .0022, BH-FDR q = .011), and sequence-order agreement was higher at both five years (d = 0.217, q = .034) and three years (d = 0.219, q = .036) before the outcome. Exploratory signals for question pressure, leap nonlinearity, and discovery readiness showed larger effects (d ≈ 0.37–0.53, q = .0012 across several offsets), but are treated as hypothesis-generating. Predictively, however, a conventional activity/citation baseline achieved ROC-AUC = .807 (95% CI [.765, .846]), exceeding the stage-only model (AUC = .713) and the book-wide scalar score (AUC = .521). Adding the book-wide score or core geometry did not improve baseline discrimination; legacy PBS produced a small, statistically uncertain increment (ΔAUC = .0146, 95% CI [−.0085, .0372]). Thus, the evidence supports a narrower conclusion than “breakthrough prediction”: future-emergent fields exhibit reproducible structural organization years before emergence, but those structural signals do not yet outperform conventional bibliometric activity for classification. The study provides a falsifiable computational operationalization, independent validation, negative-result transparency, and a fully reproducible OpenAlex/PubMed software pipeline under Apache-2.0.