For further validation, we compare a model SFT'ed only on binding, vs. end-to-end SFT and RL. Binding-aware models are extremely cost effective, and can outperform end-to-end trained pipelines.