Semi-circular bunds are widely built to rehabilitate degraded drylands, but their performance in rangelands receiving only ~100 mm rain per year is poorly documented. We evaluated hand-built semi-circular bunds (3 m diameter, 30–40 cm deep, 70–80 bunds ha⁻¹) 11–12 years after construction (built in 2005; surveyed 2016–2017) in the Chahgurik rangeland, Mirjaveh, southeast Iran, an Artemisia sieberi steppe at 1800 m elevation. Canopy cover was measured in ten 2 m2 plots in a bunded area and an adjacent unbunded control, and soil N, P and K were determined under shrubs and on bare soil at 0–10 and 10–40 cm. Mean canopy cover was 8.77% in the bunded area and 7.73% in the control, a difference of 1.04 percentage points (95% BCa bootstrap CI 0.21–1.99; exact permutation p = 0.043) or +13.5% relative (95% CI 1.1–27.5%). The standardised difference was large (Hedges g = 0.93) but imprecise, with power of only 0.51; about 19 plots per area would be needed for 80% power. Available P and K were lower in the bunded area at both depths and positions, whereas total N was higher only in the surface soil under shrubs. A runoff-capture analysis bund geometry showed a catchment-to-basin ratio of about 37:1, far above 1.4:1–3:1 of FAO reference designs. Under an assumed runoff coefficient of 0.10–0.50 (reference 0.25) and an assumed mean rain-event depth of 4–12 mm, neither measured at site, each catchment generates a runoff-equivalent depth of 10–50 mm yr−1. Annual water depth entering each basin is estimated at 7.2 times 100 mm rainfall depth (median; 90% range 3.7–10.9 from Monte Carlo propagation), with runoff coefficient accounting for most uncertainty. The 0.7–1.0 m3 basin storage captures most run-on, but basins cover <3% of the land. The bunds thus redistribute water rather than add it, consistent with modest area-level vegetation difference. Because the comparison involves one bunded and one control area without pre-construction data, observed differences indicate an association with bunding rather than a demonstrated treatment effect. In such dry rangelands, bund density and basin size should be designed with area expected to respond, and monitoring should use replicated, baseline-referenced designs.