foreman-stewardship-interrupted-tick

Answer

Problem restated. Tick 129's worker completed all of RR-VIS-002C's work (worker commits a3fbe09+f89adc8, judge PASS 623d59f5, screenshots refreshed 22:30) but died PRE-COMMIT mid-bookkeeping: it appended only dispatch event 218, never bumped the header (ticks_total stuck at 128), never marked the tasks.jsonl row complete, and committed nothing. Recovery posture: adopt the dead tick's finished work; never respawn the worker (a respawn would duplicate the judge run and corrupt the event ledger).

Fix — 6 steps.

Step 1 — Independently verify every claim before touching the board. No trust, no git log --grep. Each gate is a separate command so a single false claim is caught in isolation:

git log --oneline -3                        # expect a3fbe09, f89adc8 present
npm run typecheck                           # clean
npm run validate:configs                    # expect 11+10 = 21 valid
npm run unit                                # expect 678/678
npm run verify:dist                         # dist builds from HEAD
git show 623d59f5 --stat                    # judge verdict PASS on disk
stat -c '%y' e2e/screenshots/*.png          # mtimes == 22:30, unregenerated

Step 2 — Append board events 219+220 and bump the header with append_board_task_completed.py (guard rails make it safe to run exactly once and idempotent on re-run):

# tools/append_board_task_completed.py
import json, pathlib, datetime

BOARD  = pathlib.Path("dispatch/board.json")
TASKS  = pathlib.Path("dispatch/tasks.jsonl")

def main() -> None:
    board = json.loads(BOARD.read_text())
    ev = board["events"]; hdr = board["header"]

    # --- preflight: reject any drift before mutating ---
    assert ev[-1]["id"] == 218, f"unexpected tail id={ev[-1]['id']}, abort"
    assert hdr["ticks_total"] == 128, f"header already at {hdr['ticks_total']}, abort (idempotency guard)"
    assert not any(e["id"] in (219, 220) for e in ev), "events 219/220 already exist, abort"

    # --- adopt the dead worker's finished work ---
    ev.append({"id": 219, "kind": "task_completed", "detail": {
        "task": "RR-VIS-002C",
        "work_commits": ["a3fbe09", "f89adc8"],
        "judge_verdict": "PASS", "judge_commit": "623d59f5",
        "screenshots_refreshed_at": "22:30"}})
    ev.append({"id": 220, "kind": "board_bookkeeping_complete",
               "detail": {"placeholder": True}})   # full audit object patched in Step 3
    hdr["ticks_total"] = 130                        # 128 -> 130 = two adopted events

    # --- mark the tasks.jsonl row complete exactly once ---
    rows = [json.loads(l) for l in TASKS.read_text().splitlines() if l.strip()]
    for r in rows:
        if r.get("task") == "RR-VIS-002C":
            assert not r.get("complete"), "row already complete, abort"
            r.update({"complete": True, "judge": "PASS 623d59f5",
                      "commits": ["a3fbe09", "f89adc8"], "completed_at": datetime.datetime.utcnow().isoformat()})
    TASKS.write_text("\n".join(json.dumps(r) for r in rows) + "\n")
    BOARD.write_text(json.dumps(board, indent=2) + "\n")

Step 3 — Patch event 220's detail in BOTH stores with the full audit object (load-modify-store; the tick-83/88 technique — never mutate in place via a partially-applied patch). The board and the evidence/audit store are separate; both must carry the identical audit payload:

# tools/patch_event_220_audit.py
AUDIT = {
    "task": "RR-VIS-002C",
    "adopted_from_tick": 129,
    "adopted_by": "foreman stewardship (never respawned worker)",
    "work_commits": ["a3fbe09", "f89adc8"],
    "judge_verdict": "PASS", "judge_commit": "623d59f5",
    "screenshots_refreshed_at": "22:30",
    "board_events_appended": [219, 220],
    "ticks_total_after": 130,
    "tasks_row_complete": True,
}

def patch(store: pathlib.Path) -> int:
    data = json.loads(store.read_text())
    n = 0
    for e in data["events"]:
        if e["id"] == 220:
            e["detail"] = AUDIT; n += 1
    if n: store.write_text(json.dumps(data, indent=2) + "\n")
    return n

assert patch(BOARD) == 1 and patch(EVIDENCE_STORE) == 1   # BOTH stores, not one

Step 4 — Parity probe. The two stores must agree event-for-event, id-for-id, detail-for-detail:

def parity(a: dict, b: dict) -> None:
    ea, eb = {e["id"]: e for e in a["events"]}, {e["id"]: e for e in b["events"]}
    assert ea.keys() == eb.keys(), "event id sets differ"
    assert all(ea[i] == eb[i] for i in ea), "detail drift between stores"
    assert a["header"]["ticks_total"] == b["header"]["ticks_total"] == 130
    print(f"PARITY MATCH {len(ea)}/{len(eb)} max {max(ea)}")   # expect: MATCH 215/215 max 220

Step 5 — Two commits, co-author trailer as the second -m (trailer must be its own -m so it lands in the message body, not the subject):

git add dispatch/board.json dispatch/tasks.jsonl
git commit -m "chore(dispatch): adopt RR-VIS-002C bookkeeping (events 219-220, ticks 128->130)" \
           -m "Co-authored-by: Foreman Steward <<email>>"        # -> 746393e
git add dispatch/evidence
git commit -m "chore(evidence): record RR-VIS-002C verdict/evidence audit" \
           -m "Co-authored-by: Foreman Steward <<email>>"        # -> 9f99c29

Step 6 — Confirm terminal state. Board: 45 complete / 3 pending, ticks_total == 130, event tail ...218, 219, 220, parity MATCH 215/215 max 220.