The graphlemon Work Node (planner) workflow finished. It produced a benchmark/test dataset artifact: **deploy_stage_concurrency_mutex_status_matrix** — a matrix mapping every `deployment.status` enum value (from `src/schema.py`'s CHECK constraint) to the expected allow/reject outcome of a subsequent `stage_deployment()` call, derived from the `ux_deployment_active` partial unique index's exclusion set (`status NOT IN ('finalized', 'rolled_back', 'blocked_pending_human')`).

Key points from the result:
- Only 3 prior statuses allow a new stage: `finalized`, `rolled_back`, `blocked_pending_human` (plus the bootstrap case of no prior row).
- All other statuses — including failure-ish ones like `health_check_failed` and `rollback_failed`, which are *not* in the index's exclusion set — must still reject with `ConcurrentDeploymentError`, not a bare `sqlite3.IntegrityError`.
- Additional assertions: rejected calls must not insert a new `deployment` row or launch any candidate checkout/subprocess; allowed calls must produce the existing `staged -> candidate_starting` progression and return shape.

No files were modified — this was a planning/dataset-generation artifact. Let me know if you'd like this turned into an actual test file or artifact commit.