This paper presents a method for dynamically assessing parametric variable importance and likely influence on performance objectives as a large, precomputed design space is filtered down to explore more specific problems. Custom parametric models coupled with performance simulation can support early design, but they can be inflexible and are not always created in practice due to time and other constraints. Large parametric datasets of previously simulated design subproblems could thus make performance-based modeling more accessible, but they can have too much information and fail to focus on supporting design decisions for specific variables and ranges. Using a parametric daylight room model as an example, we first train a linear model tree. As variable bounds are filtered and adjusted by a designer, remaining coefficients are interpolated to provide an adjusted variable importance for the new domain.