In mountain watersheds under development pressure, separating climatic from anthropogenic shares answers only half of the question; the other half is through how many pathways, and how visibly, the human driver acts. A two-level attribution framework was applied to the 798.1 km2 Taleqan watershed. At the first level, the two shares were separated for annual streamflow decline, by climate elasticity on sixty-year records of six gauging stations, and for the 2050-horizon peak discharge, by a twelve-combination matrix of a nine-sub-basin rainfall-runoff model in a randomised complete block design. At the second level, a two-by-two factorial design split the human driver into a visible lever, settlement expansion, and a hidden lever, rangeland condition loss without class change; land use was classified for five epochs, 2005 to 2025, with Random Forest into six classes, two being qualitative rangeland sub-classes. The human driver dominated both components. At the outlet, 68 per cent of the 17.9 per cent annual-flow decline is non-climatic, with 30 to 77 per cent across the six stations. For the 2050-horizon 100-year peak, the isolated land-use role is 11.6 per cent against 7.9 per cent for climate, so about 60 per cent of the attributable change is anthropogenic. The climatic share, however, depends entirely on the monthly window of the precipitation change factor, ranging from -2 to +314 per cent of baseline. At the second level, 96 per cent of the curve-number range originates from the hidden lever and only 4 per cent from the visible; bootstrapping kept the non-climatic share between 58 and 79 per cent. The vegetation index does not reveal this lever, because the warming signal dominates it; specific erosion is a suitable alternative. Reconciling development with conservation therefore requires grazing reform as the primary instrument, supported by adaptation to heavier rainfall.