<p>While citation counting has long been the standard for retrospectively operationalizing and assessing high-impact research due to its objectivity, its utility for prospective prediction is limited by accumulation time lags and structural biases (e.g., the rich-get-richer effect). To overcome these constraints and enable the early forecasting of high-impact outcomes, the field has evolved toward incorporating alternative signals and advanced modeling techniques. This review traces the evolutionary trajectory of high-impact prediction research. We first examine the foundational role of citation analysis as both a target definition and a baseline heuristic. We then explore how researchers have integrated early-available alternative sources (e.g., journal attributes, author reputation, and content signals) to bridge the gap before citations accumulate. Furthermore, we analyze the progression of methodological approaches, ranging from network analysis and regression models to the recent advances in machine learning. By distinguishing between identification and prediction tasks, this review provides a comprehensive historical synthesis of how the field has advanced from simple metrics to complex predictive systems.</p>