This is indeed a strange problem, and my first guess would be that you have variables that RMD can access but are not correctly set during knitting, but since you already cleared the output and restarted R that should have made the markdown break too...
In order for us to help you with your question, please provide us a minimal reproducible example where you provide a minimal (dummy) dataset and code that can recreate the issue. Once we have that, we can go from there. For help on creating a Reprex, see this guide: