{ "background": "Industrial machinery fleets in developing economies face significant operational risks, yet there is a scarcity of high-fidelity field data to evaluate risk reduction methodologies. Existing studies often rely on retrospective analyses or self-reported data, limiting the robustness of evidence for engineering interventions.", "purpose and objectives": "This data descriptor presents a novel, randomised field trial dataset designed to methodologically evaluate a proactive maintenance and operator training protocol. The primary objective was to generate empirical evidence on the efficacy of structured interventions for reducing machinery downtime and safety incidents.", "methodology": "A randomised controlled trial was conducted across multiple industrial sites. Fleets were randomly assigned to intervention or control groups. The intervention group received a bundled protocol of scheduled condition-based maintenance and simulator-based operator training. Data were collected over a trial period on machine availability, incident reports, and component failure rates. The core analysis employs a generalised linear mixed model: $\\log(E(Y{it})) = \\beta0 + \\beta1 Ti + \\beta2 X{it} + ui + \\epsilon{it}$, where $Y{it}$ is the monthly downtime hours for machine $i$ at time $t$, $Ti$ is the treatment indicator, $X{it}$ are time-varying covariates, and $ui$ is a machine-specific random effect.", "findings": "The dataset comprises 1,247 machine-months of observations. Preliminary analysis indicates a reduction in unplanned downtime in the intervention group, with a risk ratio of 0.72 (95% CI: 0.64 to 0.81) relative to the control. Uncertainty in the estimate was quantified using robust standard errors clustered at the site level.", "conclusion": "The dataset provides a rigorous empirical foundation for evaluating engineering risk management strategies in an industrial context. It captures the real-world implementation and heterogeneous effects of a combined technological and human-factor intervention.", "recommendations": "Researchers should utilise this dataset for further analysis of interaction effects and cost-ben