Energy Performance Certificates (EPCs) are a central instrument of building energy policy in the European Union, yet their reliability depends critically on the consistency of the underlying assessment methodology. In Germany, two legally equivalent but methodologically distinct certificate types coexist: demand-based certificates, which calculate theoretical energy demand under standardized conditions, and consumption-based certificates, which are derived from metered energy consumption data. This paper provides the first large-scale empirical comparison of the two certificate types for the German residential building stock, using Coarsened Exact Matching (CEM) to construct observationally comparable groups of buildings differing only in certificate type. Drawing on approximately 392,700 matched property listings from the RWI-GEO-RED dataset (2007–2025), we estimate a conditional demand–consumption gap of 44.5 kWh/m²a, or approximately 29.4% of the consumption-based mean. The gap varies systematically with building vintage – exceeding 50% for pre-1958 buildings and reversing to -16% for post-2015 buildings – and with heating technology. Quantile regression confirms that discrepancies are present throughout the efficiency distribution and are largest among the most energy-inefficient buildings. Misclassification analysis reveals that 77.9% of comparable building pairs receive different energy efficiency class assignments depending on certificate type, with 28.6% straddling the F/G boundary at 200 kWh/m²a – a threshold of particular salience for owners, buyers, and policymakers assessing the worst-performing segment of the residential building stock. These findings call into question the legal equivalence of the two certificate types and have direct implications for the transparency of real estate markets and the consistency of building energy regulation.