I derive a condition on the difference between two gravitational waveforms that is sufficient to guarantee that they will furnish equivalent parameter estimates in a Bayesian analysis. The condition is that the log-likelihood difference between the two waveforms at the same point in parameter space is small. I show that this condition is equivalent to the difference between the two waveforms being small compared to the signal-to-noise ratio of the waveforms; the condition does not require that either waveform be a good model for any signal in the data.
TODO: Read Hu & Veitch (2022) carefully which I think derives an essentially identical criterion from a similar argument.