This study employs a marginal returns framework using WitikaStats to analyze the non-linear relationship between direct taxes and economic growth in Zambia (1990-2024). Moving beyond insignificant average regression results, the analysis identifies distinct returns regimes: negative at low shares (31.3%-35.6%), increasing (35.6%-43.7%), diminishing (43.7%-51.8%), and negative again beyond 51.8%. A precise growth-maximizing optimal direct tax share is identified at 51.83%, with an early warning inflection point at 43.72%. The findings provide actionable, data-driven fiscal policy guidance, recommending the government maintain the direct tax burden within a 43.7%-51.8% band of total revenue. This demonstrates how advanced computational economics can operationalize theoretical concepts like diminishing returns to solve real-world policy problems, offering a replicable model for evidence-based fiscal governance. As such, this approach transforms marginal analysis from a classroom abstraction into a vital tool for solving real-world problems, enabling evidence-based fiscal strategy that dynamically responds to an economy’s position on its returns frontier, thereby unlocking higher and more stable growth.