Yield stability in rainfed smallholder systems is governed as much by the chemical composition of staple crops as by agronomic practice, yet food chemistry is rarely treated as a modifiable input to agricultural adaptation. The design exploits the staggered rollout of a national agricultural extension programme that promoted chemistry-informed variety selection across districts between 2015 and 2021. We specify a two-way fixed-effects estimator with district and year fixed effects, and we address treatment-effect heterogeneity through a staggered adoption framework. The analysis is designed to identify whether districts adopting chemistry-adapted varieties experienced reduced yield variance relative to non-adopting districts, controlling for rainfall, soil fertility, and market access. We propose a composite spatial index integrating slope, drainage density, rainfall intensity, land cover, soil permeability, elevation, and distance to channels to delineate treatment-relevant agroecological zones. The framework clarifies identification assumptions, proposes robustness checks, and discusses threats including non-random programme placement and spillover effects.