Transport and economic geographers' accessibility models provide nuanced descriptions of accessibility gains to investments in structured environments but struggle to accurately reflect the complex mobility constraints in rural, mountainous areas of the developing world. This forces planners in such contexts to employ cruder measures of accessibility gains to each road, particularly where large collections of feeder roads inflate data collection expenses. To address this disconnect we develop a scalable method for evaluating the cost-efficiency of rural roads investments based on seasonal accessibility improvements to specified services. Accessibility improvements are measured using road-specific multi-modal cost-distance models, incorporating terrain, seasonal effects, and extensive off-network walking travel. Using our models, we estimate and compare accessibility improvements for a large collection of proposed rural feeder roads in a case study of Nepal's remote, mountainous Karnali province. The developed model and workflow can be adapted to compare accessibility gains from roads or other investments in equivalently rugged, remote, and data-poor environments, opening the door to more rigorous accounting of accessibility in a traditionally neglected context.