We propose a new method for climate-based daylight modeling (CBDM) based on simulating and evaluating only the most important features. By adaptively sampling the temporal lightfield that describes daylight in buildings, our method escapes the curse of dimensionality that binds typical approaches. The method is centered around an iteratively guided sampling strategy that produces unordered sky and sun coefficient samples throughout an architectural space. This data is organized into a lightfield data structure that can be evaluated under any sky condition. Outputs include photometric quantities, visual comfort metrics, and high dynamic range images. Unlike existing CBDM methods, our approach is not limited by the rendering algorithm, material model, level of detail, or metric output. Our method is accurate and fast. As such it can serve as a general interface for any view or point-based daylighting question, including those answered by illuminance and luminance based metrics.