How does a CI eval job avoid stacking a new comment on every push?

It edits its own prior comment in place instead of posting a new one on every push. Every comment Tessary’s CI gate renders carries a stable marker embedded in the body, and the sample GitHub Action wires that up to a comment-upsert step, matched by comment author, that replaces the existing comment’s text rather than adding a new one below it.

Each push to the PR still triggers a fresh report from the server, but the thread only ever shows the latest verdict, not a growing stack of one per commit. This depends on the CI job actually using an upsert action rather than a plain “create comment” call. The marker exists so a job can look up its own prior comment and know it’s the one to replace, but the job still has to do that look-up; nothing on the platform side enforces it.