In the case of MR-SAD, how bad can model bias be? Does it also depend on the data resolution and resolution at which there is still enough anomalous signal? Or can the anomalous signal help in overcoming the bias inherent in the MR model?