> Feedback should almost always be based on data

I do agree with you here. What I wanted to express is that you should hold back on interpretation. If you talk about your own perceptions, it allows other interpretations to also be valid.

Thank you for the example on clogging the CI servers. Indeed, I would find it hard to turn this into positive reinforcement. Also I would agree that it's dishonest in not telling them. I guess I would try to highlight my own pains with their behavior like "I was blocked five times in deploying because of your failing tests. Please run your tests locally first.". I would always want to give them the opportunity to take my perspective and understand my reasoning.