Basel III credit-to-GDP gaps are used to assess whether aggregate credit is excessive or not and inform macroprudential policymaking. Yet, estimates from Basel III’s prescribed detrending procedure are prone to continuous reevaluations that do not reflect changes in the data and exceed commonly discussed end-of-sample biases from converging one- and two-sided filtering procedures. To illustrate the extent of unreliability, I introduce historical reliability bands. Based on simulation and empirical evidence for 43 countries, I show that estimates do not converge to full sample estimates and each quarter is associated with a new trend history, which compromises the comparability of cyclical positions over time. This leads to relevant misalignment in countercyclical buffer decisions in both, direction and size, and may impair the regulator’s credibility. Alternatively using a two-year difference filter would provide endpoint and historically reliable estimates, while yielding distinctly fewer and smaller decision misalignment and maintaining credit gap dynamics and amplitudes that the regulator deems important.