sprint_orchestration_recoverlisted

# sprint_orchestration_recover Recover one stalled transition without replaying completed work. ## Establish the last durable truth Read: ```sh ./sc sprint board --sprint <doc-id> ./sc sprint status ./sc sprint alerts ./sc mem message check ./sc job list --all ./sc watch list --all ``` For the affected unit, confirm: - last scoped task and result; - board state, branch, PR, developer, and reviewer; - live process or job evidence; - PR head, checks, review state, and merge state; - binding batch, park, quarantine, and alert state. Quiet is a symptom. Declare a stall only when an expected transition is absent and positive evidence shows no active process, job, review, CI run, or queued wake that can produce it. Validate every absence instrument against a known-positive target before using its empty output. A missing command, rejected predicate, non-zero probe, or positive control it cannot see makes the claim unmeasured, not absent. Inspect the probe's exit status and stderr; never translate empty stdout alone into "nothing is running." A reconciler finding requests a checkup. For a headless shell: 1. Resolve the assigned subject from the sprint board, scoped task, launching planner, and worktree. 2. Find the harness process whose `/proc/<pid>/cwd` is that worktree. 3. Read `/proc/<pid>/stat` and reject state `Z`; a zombie directory and stat file are not liveness. 4. Sample `/proc/<pid>/stat` twice over a bounded interval and compare `utime + stime`. 5. Treat