<p>Creep ringing is typically analyzed using canonical models like the Kelvin-Voigt or Jeffreys models. Such analysis is used to remove inertial artifacts and estimate model parameters, viscoelastic moduli, and creep compliance <i>J</i>(<i>t</i>). We investigate the accuracy of the <i>J</i>(<i>t</i>) inferred by such analysis for materials with nontrivial retardation spectra. To this end, we introduce Glacier – a program that predicts how inertia modifies <i>J</i>(<i>t</i>). We generate synthetic data with generalized linear models and fit them with canonical models to study model-specification error. We find that the error in fitting data is correlated with the error in predicting <i>J</i>(<i>t</i>), although the latter is larger and widely dispersed. Application to the Rouse model illustrates that a visually good fit to ringing oscillations does not guarantee accurate recovery of <i>J</i>(<i>t</i>) beyond the fitting window or viscoelastic moduli far from the ringing frequency.</p>