The merge job deleted the PR branch unconditionally after attempting the squash merge, even when the merge itself failed (e.g. HTTP 405). That's why PR #8 closed with the branch gone but no commit landed - the merge failed and cleanup ran anyway. Now the branch is only deleted after a confirmed 200/201, and the job exits non-zero on any other outcome so it's visible in the run status instead of silently continuing. Also add a concurrency group, since update/build/push/schedule/PR triggers overlapping could race two merge attempts against the same PR.