Hi again
Looking through the code again in detail I found the solution.
When embedding the ‘predicted’ TX timestamp in the final message, I forgot to zero the lower 9 bits. This meant there were a discrepancy between the actual time the delayed message was transmitted and the predicted TX timestamp, which caused the noise that you see in my plots.
Now that it is fixed, I get the expected accuracy of the ranging measurements. However I have some funny sudden outliers that I can’t fully explain at this point.
Thanks again Andy for your response.