Estimates of noise channels for quantum gates are required for most error mitigation techniques and are desirable for informing quantum error correction decoders. These estimates can be obtained by resource-intensive off-line characterization techniques, but can become stale due to device drift and fluctuations. We propose a method to address this issue by performing real-time adaptation of noise channel estimates during the execution of a quantum algorithmic circuit using extended flag gadgets, mid-circuit measurements and Bayesian inference. We carry out analytical calculations and numerical simulations employing a Dirichlet prior distribution for the error rates in a Pauli channel to demonstrate and evaluate the technique, which can be seen as a protocol for real-time calibration of high-level gate error information.