Which envelope inputs deserve verification first? An analytical Sobol study of an anonymised nonresidential building inventory

Detailed building reports can contain hundreds of envelope entries but little information about the reliability of the repeated thermal properties. This study examines how the assumed structure of input uncertainty changes the priorities for verifying such an inventory. A confidential baseline-only energy report is used to construct a normalised conductance inventory from areas and thermal transmittances. Seven source-defined component groups are analysed using exact first-order and total-order Sobol indices for an additive model. Two deliberately illustrative uncertainty specifications are compared: independent group multipliers uniformly distributed within ±10%, and a wider ±30% interval for the door group while other intervals remain unchanged. Under equal relative uncertainty, the wall group contributes approximately 33.55% of output variance, followed by roofs at 20.74% and windows at 18.57%. Widening door uncertainty changes the leading group to doors, with 53.16% of variance. Assuming independent row errors rather than shared group errors reduces the calculated standard deviation by a factor of approximately 3.97, despite using the same marginal uncertainty width. The resulting indices describe uncertainty in a screening inventory, not annual energy demand or renovation savings. Boundary temperature corrections, ground coupling, thermal bridges and ventilation are excluded explicitly. The contribution is a transparent demonstration of how verification priorities depend jointly on nominal conductance, uncertainty width and error dependence. All equations and the anonymised aggregate inputs are given in the paper, so that every reported number can be recalculated without disclosing the building identity.