A reproducible run-level workflow for internal–external training-load integration in alpine skiing

Monitoring training load in alpine skiing requires physiological and mechanical information to be aligned under difficult field conditions. Proprietary ecosystems often reduce this problem to a single summary score, but their black-box design limits transparency and makes sensitivity analysis difficult. The present proof-of-workflow and reproducibility demonstration uses a table containing 14 downhill runs to examine the interpretive behavior of a transparent and reproducible post-segmentation analytical workflow. It is not intended as a population-level physiological validation, reliability assessment, or deployment study. The engineering contribution is the preservation of intermediate outputs and inspectable weighting and comparison diagnostics, rather than new sensing technology, signal processing, or a mathematical fusion model. A deterministic alternative-scaling sensitivity analysis compared the supplied within-dataset z values, median/MAD scaling, and centered empirical-rank scaling. Median/MAD scaling preserved the CL(0.5) ranking, whereas centered empirical-rank scaling retained high ordinal agreement with the supplied-z baseline ( ρ = 0.961) but changed one member of the top-3 set; alternative scaling could also change the grid-based balance point. At α = 0.50, the fused score correlated equally with the internal and mechanical components ( r = 0.965 for both). This is a dataset-specific equal-correlation diagnostic, not an optimal or recommended physiological, engineering, or coaching weighting. Within this 14-run demonstration dataset, the fused score retained complementary information across the two components. The workflow supports transparent reporting and reproducible comparison.