Despite being one of the most cost-effective and sustainable modes for transporting freight, railways globally have been steadily losing market share, in large part due to the slow speed of freight trains. For example, in Indian Railways, the world’s fourth-largest network, the average freight speed has stagnated at around 25 kmph for decades. An important operational friction is mixed passenger-freight operation: scheduled passenger services can receive priority from section controllers, interacting with queue spillovers and capacity constraints. This paper studies the causes of freight delays by modeling how section controllers make stop-and-hold decisions for freight trains. Using high-frequency network-level data, we estimate an empirical network decision model of section controllers’ decision making that incorporates spatial correlation in unobservables via a copula-based approach and addresses endogeneity through a network-based instrumental variables strategy. The estimated model quantifies key operational margins, such as queue spillovers, prioritizing passenger trains, and strategic holding at large stations, and enables simulated counterfactuals. We use the model to evaluate interventions aimed at improving freight performance, including both infrastructure-heavy (Freight-Only Corridors) and operational (e.g., time-based dispatch thresholds and vertical consolidation of freight capacity) alternatives. To study the effect of a Freight-Only Corridor (FOC), we simulate a benchmark environment with no passenger traffic on the corridor. This yields a 43% reduction in dwell time and a 15% increase in freight speeds. We further establish conservative bounds on these benefits by incorporating additional design enhancements, such as fewer stations and higher inter-station speeds, motivated by the government’s planned Dedicated Freight Corridor. In addition, we assess two low-cost operational interventions: time-based threshold dispatch and vertical stacking of freight. We find that these non-capacity interventions can deliver benefits comparable to those of FOCs while being considerably cheaper. Specifically, a 60-minute threshold rule increases speeds by 24%, while stacking trains to consolidate 30\% of freight capacity raises speeds by 7% in the baseline analysis. We also account for real-world implementation frictions associated with these interventions, including interference with scheduled passenger trains and time losses from train consolidation. Overall, our paper offers a rich, actionable, and policy-relevant menu of infrastructure and operational interventions for improving freight movement.