Randomized field trials have been used to evaluate smallholder farming systems in Tanzania with mixed results, highlighting the need for methodological replication studies. A detailed design was employed to replicate a previous trial, including random assignment of plots, equal distribution of inputs (fertilizers, seeds), and data collection over multiple seasons. Statistical analysis utilised a linear mixed-effects model to account for the hierarchical structure of the data. In one replicated field trial, an increase in crop yield was observed with fertilizer application compared to control plots, although this effect varied significantly across different soil types (p < 0.05). The replication study confirmed the robustness of the original findings and underscored the importance of considering site-specific conditions for optimal intervention effectiveness. Future studies should incorporate more detailed data on soil properties to enhance predictive accuracy, particularly in regions with diverse climatic zones. The empirical specification follows $Y=\beta_0+\beta^\top X+\varepsilon$, and inference is reported with uncertainty-aware statistical criteria.