I have noticed that some CNOT gates (for instance, between qubits 1 and 2 in
ibmq_casablanca) display an error rate of 1 while typical error rates are of course much smaller. How is that even possible? What does it mean?
The devices are calibrated daily(ish) and sometimes a bad calibration requires a "rerun". The value 1 is some sort of flag that something went wrong with the calibration.
Noise-aware transpilation algorithms will try to avoid that gate as will be presented as particularly noisy.
Thanks @paul-nation for the help answering this question!