Fuzzy automated reviews should always run in an interactive loop with a developer on their workstation and contain enough context to quickly assess if they are valid or not.
When developers create a PR, they already feel they are "done", and they have likely already shifted their focus on another task. False positive are horrible at this point, especially when they keep changing with each push of commits.
When developers create a PR, they already feel they are "done", and they have likely already shifted their focus on another task. False positive are horrible at this point, especially when they keep changing with each push of commits.