As jurisdictions and instrument developers look to optimize scoring to meet specific population needs for risk assessment, an open question is which scoring method is most optimal. Popular methods range from manual simple scoring approaches (e.g. Burgess) to more complex machine learning algorithms (e.g. random forests) to develop optimized scoring. Prior comparisons between scoring approaches have shown that different methods appear to produce similarly acceptable levels of predictive validity (Cunningham & Sorensen, 2006; Ghasemi et al., 2020; Grann and Långstrӧm, 2007; Hamilton et al., 2014; Leu et al., 2011; McLaren & Rieger, 2024; Tollenaar & van der Heijden, 2013). Considerations for item inclusion and weighting of static over dynamic information is also important depending upon what the tool is designed to inform. The present research provides a comparison of scoring methods across prediction as well as an overview of the drawbacks of each approach. Scoring was developed for the ReduCE (Author, 2024) tool using manual (Burgess, Nuffield, modified Nuffield, regression) and machine learning (artificial neural network, random forests) with an unweighted comparison. ReduCE intentionally incorporates items to gauge current risk, including historic static criminal behavior, recent behavior, and evidence of strength factors that may mitigate risk. To mimic parole practice, the optimal scoring approach ensures item inclusion and static item weight does not negate the impact of information that inform recent behavior and may mitigate risk. Overall, the machine learning approaches did not outperform the manual methods in this study. Use of machine learning methods did not outweigh the threats to decision-making introduced by the lack of transparency in item inclusion and weighting.