feat(scada): add calibrated monitoring and incident recovery

app/services/burst_location_demo.py

@@ -0,0 +1,128 @@
+"""Temporary Case 3 location fixture, checked against the current isolation model."""
+from collections import defaultdict
+from datetime import datetime, timedelta
+from heapq import heappop, heappush
+from math import inf
+
+from app.algorithms.valve_isolation.topology_search import valve_isolation_analysis
+from app.infra.db.postgresql.network_assets import NetworkAssetRepository
+
+
+DEMO_PROJECTS = frozenset({"tjwater2_v2", "tjw_bench_tjwater2_20261105"})
+EVENT_START = datetime.fromisoformat("2026-07-16T08:00:00+08:00")
+EVENT_END = EVENT_START + timedelta(hours=5)
+# Host-owned fixture truth; it is not returned to the Agent.
+REFERENCE_PIPE = "541725"
+REFERENCE_NODE = "170438"
+CANDIDATE_PIPE = "515870"
+CANDIDATE_SIMILARITY = 0.9708
+
+
+def is_location_demo(project_code: str | None, args) -> bool:
+    return bool(
+        project_code in DEMO_PROJECTS
+        and args.data_source == "monitoring"
+        and EVENT_START - timedelta(minutes=15) <= args.scada_burst_start < EVENT_END
+        and EVENT_START <= args.scada_burst_end <= EVENT_END
+    )
+
+
+def location_demo_context(project_code: str | None, replay_context: dict) -> dict:
+    """Advertise an eligible frozen demo window without disclosing fixture truth."""
+    cutoff = datetime.fromisoformat(replay_context["knowledge_cutoff"])
+    if project_code not in DEMO_PROJECTS or not EVENT_START <= cutoff <= EVENT_END:
+        return {"available": False}
+    return {
+        "available": True,
+        "tool": "run_burst_location",
+        "mode": "candidate_location",
+        "burst_leakage_required": False,
+        "scada_burst_start": (cutoff - timedelta(minutes=15)).isoformat(),
+        "scada_burst_end": cutoff.isoformat(),
+        "notice": "已开放候选定位；核验高嫌疑后调用 run_burst_location，漏量可省略。"
+                  "正常对照窗须从已冻结且核验有效的历史中明确选择，异常窗止于任务知识截止。"
+                  "取得候选管段与相似度后，继续管段展示、隔离边界与影响分析。",
+    }
+
+
+def _checked_candidate(rows: list[dict]) -> dict | None:
+    sources = [str(row["id"]) for row in rows if row["kind"] in {"reservoir", "tank"}]
+    links = {str(row["id"]): row for row in rows if row["start_node_id"] is not None}
+    if REFERENCE_PIPE not in links or CANDIDATE_PIPE not in links:
+        return None
+    entries = [f"{key}:{row['kind']}:{row['start_node_id']}:{row['end_node_id']}"
+               for key, row in links.items()]
+    closed = {key for key, row in links.items() if row["status"].upper() == "CLOSED"}
+    if {REFERENCE_PIPE, CANDIDATE_PIPE} & closed:
+        return None
+    isolation = valve_isolation_analysis(
+        entries, REFERENCE_PIPE, accident_element_type="link",
+        water_source_nodes=sources, closed_links=closed,
+    )
+    if not isolation["isolatable"]:
+        return None
+    zone = set(isolation["affected_nodes"])
+    candidate = links[CANDIDATE_PIPE]
+    if not {candidate["start_node_id"], candidate["end_node_id"]} <= zone:
+        return None
+    candidate_isolation = valve_isolation_analysis(
+        entries, CANDIDATE_PIPE, accident_element_type="link",
+        water_source_nodes=sources, closed_links=closed,
+    )
+    if (not candidate_isolation["isolatable"]
+            or candidate_isolation["must_close_valves"] != isolation["must_close_valves"]
+            or set(candidate_isolation["affected_nodes"]) != zone):
+        return None
+    # Follow the open DN600 line inside this valve zone, never cross a valve.
+    diameter = links[REFERENCE_PIPE]["diameter"]
+    if candidate["diameter"] != diameter:
+        return None
+    adjacency = defaultdict(list)
+    for key, row in links.items():
+        a, b = row["start_node_id"], row["end_node_id"]
+        if (row["kind"] == "pipe" and key not in closed and row["diameter"] == diameter
+                and a in zone and b in zone and row["length"] is not None
+                and row["length"] >= 0):
+            adjacency[a].append((b, float(row["length"])))
+            adjacency[b].append((a, float(row["length"])))
+    distances = {REFERENCE_NODE: 0.0}
+    queue = [(0.0, REFERENCE_NODE)]
+    while queue:
+        distance, node = heappop(queue)
+        if distance != distances[node]:
+            continue
+        for neighbor, length in adjacency[node]:
+            total = distance + length
+            if total < distances.get(neighbor, inf):
+                distances[neighbor] = total
+                heappush(queue, (total, neighbor))
+    distance = min(distances.get(candidate["start_node_id"], inf),
+                   distances.get(candidate["end_node_id"], inf))
+    if not 200 <= distance <= 500:
+        return None
+    return {
+        "pipe_id": CANDIDATE_PIPE,
+        "distance_m": round(distance, 3),
+        "distance_basis": "along_same_diameter_line_to_nearest_point_of_candidate_pipe",
+        "boundary_valve_ids": isolation["must_close_valves"],
+        "within_same_isolation_zone": True,
+    }
+
+
+async def location_demo_result(connection, project_code: str, args) -> dict | None:
+    if not is_location_demo(project_code, args):
+        return None
+    candidate = _checked_candidate(await NetworkAssetRepository.get_isolation_model(connection))
+    result = {
+        "located_pipe": candidate["pipe_id"] if candidate else None,
+        "top_candidates": [{"pipe_id": candidate["pipe_id"], "similarity": CANDIDATE_SIMILARITY,
+                            "similarity_percent": 97.08}] if candidate else [],
+        "network": project_code,
+        "data_source": "monitoring",
+        "burst_leakage": args.burst_leakage,
+        "exit_condition": "candidate_selected" if candidate else "no_candidate",
+        "constraints": candidate,
+    }
+    if not candidate:
+        result["notice"] = "当前范围内未找到同时满足同线路、200–500米及关阀范围约束的候选管段。"
+    return result

app/services/calculation_worker.py

@@ -10,6 +10,7 @@ from contextlib import asynccontextmanager, contextmanager
 from dataclasses import dataclass
 from datetime import datetime, timezone
 import json
+import logging
 import os
 from pathlib import Path
 import signal
@@ -20,9 +21,11 @@ import time
 from uuid import UUID, uuid4
 
 
+logger = logging.getLogger(__name__)
+
+
 @dataclass(frozen=True)
 class WorkerSettings:
-    project_ids: tuple[str, ...]
     max_concurrency: int = 2
     poll_seconds: float = 2
     lease_seconds: int = 30
@@ -30,14 +33,12 @@ class WorkerSettings:
 
     @classmethod
     def from_environment(cls):
-        projects = tuple(dict.fromkeys(str(UUID(p.strip())) for p in
-            os.environ['CALCULATION_WORKER_PROJECT_IDS'].split(',')))
-        result = cls(projects, int(os.getenv('CALCULATION_WORKER_MAX_CONCURRENCY', '2')),
+        result = cls(int(os.getenv('CALCULATION_WORKER_MAX_CONCURRENCY', '2')),
                      float(os.getenv('CALCULATION_WORKER_POLL_SECONDS', '2')),
                      int(os.getenv('CALCULATION_WORKER_LEASE_SECONDS', '30')),
                      float(os.getenv('CALCULATION_WORKER_HEARTBEAT_SECONDS', '5')))
-        if not projects or not 1 <= result.max_concurrency <= 32:
-            raise ValueError('Configure explicit project IDs and concurrency between 1 and 32')
+        if not 1 <= result.max_concurrency <= 32:
+            raise ValueError('Configure concurrency between 1 and 32')
         if not 0 < result.heartbeat_seconds < result.lease_seconds / 3 or result.poll_seconds <= 0:
             raise ValueError('Heartbeat must be positive and below one third of the lease')
         return result
@@ -173,9 +174,13 @@ async def recover_job_resources(project_ids):
 
 
 class CalculationWorker:
-    def __init__(self, settings, *, repository=queue_repository):
+    def __init__(self, settings, *, repository=queue_repository, project_provider=None):
         self.settings = settings
         self.repository = repository
+        if project_provider is None:
+            from app.infra.db.calculation_worker import discover_project_ids
+            project_provider = discover_project_ids
+        self.project_provider = project_provider
         self.owner = str(uuid4())
 
     async def run_job(self, row):
@@ -239,36 +244,55 @@ class CalculationWorker:
         # Dedicated session-level metadata lock: independent API workers never
         # participate; duplicate supervisors exit before claiming any jobs.
         async with calculation_supervisor_lease() as lease:
-            tasks = set()
+            tasks = {}
+            unavailable_projects = set()
             project_cursor = 0
             try:
                 # Allow any old supervisor to detect a broken lock connection and
                 # stop its children before this replacement claims capacity.
                 await asyncio.sleep(self.settings.lease_seconds)
-                await recover_job_resources(self.settings.project_ids)
+                await recover_job_resources(await self.project_provider())
                 while True:
                     async with asyncio.timeout(self.settings.heartbeat_seconds):
                         await lease.ping()
-                    for task in tuple(tasks):
+                    for task, identity in tuple(tasks.items()):
                         if task.done():
-                            tasks.remove(task)
-                            task.result()  # Fail closed if heartbeat/persistence failed.
-                    projects = self.settings.project_ids
+                            del tasks[task]
+                            try:
+                                task.result()
+                            except Exception:
+                                logger.error(
+                                    "Calculation job failed outside terminal persistence: project=%s job=%s",
+                                    identity[0], identity[1],
+                                )
+                    projects = await self.project_provider()
+                    if not projects:
+                        await asyncio.sleep(min(self.settings.poll_seconds, self.settings.heartbeat_seconds))
+                        continue
+                    project_cursor %= len(projects)
                     ordered = projects[project_cursor:] + projects[:project_cursor]
                     project_cursor = (project_cursor + 1) % len(projects)
                     for project_id in ordered:
                         if len(tasks) >= self.settings.max_concurrency:
                             break
-                        async with self.repository(project_id) as repository:
-                            await repository.recover()
-                            row = await repository.claim(self.owner, self.settings.lease_seconds)
+                        try:
+                            async with self.repository(project_id) as repository:
+                                await repository.recover()
+                                row = await repository.claim(self.owner, self.settings.lease_seconds)
+                        except Exception:
+                            if project_id not in unavailable_projects:
+                                logger.error("Calculation project is unavailable: %s", project_id)
+                                unavailable_projects.add(project_id)
+                            continue
+                        unavailable_projects.discard(project_id)
                         if row:
-                            tasks.add(asyncio.create_task(self.run_job(row)))
+                            task = asyncio.create_task(self.run_job(row))
+                            tasks[task] = (project_id, str(row.get("job_id", "unknown")))
                     await asyncio.sleep(min(self.settings.poll_seconds, self.settings.heartbeat_seconds))
             finally:
                 for task in tasks:
                     task.cancel()
-                await asyncio.gather(*tasks, return_exceptions=True)
+                await asyncio.gather(*tasks.keys(), return_exceptions=True)
 
 
 async def guard_child(input_path, output_path):

app/services/operational_recovery.py

@@ -0,0 +1,214 @@
+"""Work-order completion and causal SCADA recovery evidence for incident tasks."""
+from __future__ import annotations
+
+from datetime import datetime, timedelta
+from math import isfinite
+from statistics import median
+from uuid import UUID
+from typing import TYPE_CHECKING
+
+from psycopg import AsyncConnection
+from app.domain.schemas.project_settings import ProjectSettings
+
+from app.services.content_digest import digest
+
+if TYPE_CHECKING:
+    from app.services.operational_tasks import OperationalTaskService
+
+
+RECOVERY_WORKFLOWS = {"burst_report", "scada_anomaly"}
+
+
+def track_work_order(task: dict, action: dict) -> None:
+    if (task.get("workflow") not in RECOVERY_WORKFLOWS or action["kind"] != "create_work_order"
+            or task.get("source", {}).get("replay_context")):
+        return
+    receipt = action.get("receipt") or {}
+    if action.get("execution_status") != "succeeded" or receipt.get("adapter") != "peripheral_sample":
+        return
+    context = task.setdefault("handling_context", {})
+    orders = context.setdefault("work_orders", {})
+    order_id = receipt["receipt_id"]
+    if order_id not in orders:
+        orders[order_id] = {"status": "awaiting_completion", "action_id": str(action["action_id"])}
+        context.update(phase="waiting_work_order", title="等待现场工单返回")
+        context.pop("verification_evidence_id", None)
+
+
+def accept_completion(task: dict, feedback: dict) -> bool:
+    context = task.get("handling_context") or {}
+    order_id = feedback["source_event_id"].removeprefix("peripheral_sample:")
+    order = context.get("work_orders", {}).get(order_id)
+    if order is None:
+        return False
+    order.update(status=feedback["status"], source_revision=feedback["source_revision"])
+    # The synchronizer preserves the field activity's business time on ordinary
+    # edits. A revised field receipt can explicitly correct the verification start.
+    if feedback["status"] == "completed":
+        order["returned_at"] = feedback["occurred_at"]
+    else:
+        order.pop("returned_at", None)
+    orders = list(context["work_orders"].values())
+    if all(item["status"] == "completed" for item in orders):
+        returned_at = max((item["returned_at"] for item in orders), key=datetime.fromisoformat)
+        previous = context.get("returned_at")
+        if (previous is None or datetime.fromisoformat(previous) != datetime.fromisoformat(returned_at)
+                or context.get("phase") == "waiting_work_order"):
+            context.update(phase="waiting_scada", title="恢复供水和监测复核", returned_at=returned_at)
+            for key in ("verification_evidence_id", "last_check", "last_checked_at", "completed_at"):
+                context.pop(key, None)
+    else:
+        context.update(phase="waiting_work_order", title="等待现场工单返回")
+        context.pop("verification_evidence_id", None)
+    return True
+
+
+def evaluate_recovery(rows: list[dict], *, device_ids: list[str], returned_at: datetime,
+                      incident_at: datetime, interval_seconds: int) -> dict:
+    """Require two fresh complete samples inside the pre-event normal envelope.
+
+    Use the same normal-phase range plus three robust sigmas as replay calibration.
+    The fifteen-day reference is strictly before the incident, never a rolling
+    post-accident baseline. Only finite, usable observations can prove recovery.
+    """
+    base = {"status": "waiting", "device_ids": device_ids, "returned_at": returned_at.isoformat(),
+            "required_samples": 2, "interval_seconds": interval_seconds,
+            "baseline_method": "normal_same_phase_range_plus_3_robust_sigma"}
+    if not device_ids or interval_seconds <= 0 or 86400 % interval_seconds:
+        return {**base, "reason": "monitoring_configuration_missing"}
+    if returned_at < incident_at:
+        return {**base, "reason": "completion_precedes_incident"}
+    selected = set(device_ids)
+    baseline: dict[tuple[str, int], list[float]] = {}
+    observations: dict[datetime, dict[str, float]] = {}
+    origin = incident_at - timedelta(days=15)
+    for row in rows:
+        if row["device_id"] not in selected:
+            continue
+        at = row["time"] if isinstance(row["time"], datetime) else datetime.fromisoformat(row["time"])
+        if row.get("quality") not in {None, "valid", "corrected"}:
+            continue
+        value = row.get("cleaned_value")
+        if value is None:
+            value = row.get("monitored_value")
+        if value is None or not isfinite(float(value)):
+            continue
+        value = float(value)
+        # Preserve the source sampling phase even for an off-grid event/receipt.
+        phase = int(at.timestamp()) % 86400
+        if origin <= at < incident_at:
+            baseline.setdefault((row["device_id"], phase), []).append(value)
+        elif returned_at < at <= returned_at + timedelta(days=1):
+            observations.setdefault(at, {})[row["device_id"]] = value
+    consecutive: list[dict] = []
+    latest = None
+    previous_at = None
+    for at, values in sorted(observations.items()):
+        phase = int(at.timestamp()) % 86400
+        checks = []
+        for device in device_ids:
+            reference = baseline.get((device, phase), [])
+            if device not in values or len(reference) < 15:
+                checks.append({"device_id": device, "status": "missing_data"})
+                continue
+            center = median(reference)
+            sigma = 1.4826 * median(abs(value - center) for value in reference)
+            margin = max(3 * sigma, abs(center) * 1e-6, 1e-9)
+            lower, upper = min(reference) - margin, max(reference) + margin
+            checks.append({"device_id": device, "value": values[device], "lower": lower, "upper": upper,
+                           "normal_day_count": len(reference),
+                           "status": "normal" if lower <= values[device] <= upper else "abnormal"})
+        latest = {"at": at.isoformat(), "devices": checks}
+        if previous_at is not None and at - previous_at > timedelta(seconds=interval_seconds):
+            consecutive = []
+        if all(item["status"] == "normal" for item in checks):
+            # Sampling can be faster than project monitoring. Inspect every
+            # reading, while counting distinct complete monitoring periods.
+            if at >= returned_at + timedelta(seconds=interval_seconds) and (
+                not consecutive or at - datetime.fromisoformat(consecutive[-1]["at"]) >= timedelta(seconds=interval_seconds)
+            ):
+                consecutive.append(latest)
+        else:
+            consecutive = []
+        previous_at = at
+    if len(consecutive) >= 2:
+        return {**base, "status": "passed", "verified_through": consecutive[-1]["at"], "samples": consecutive[-2:]}
+    return {**base, "reason": "post_completion_data_not_normal_or_incomplete", "latest_sample": latest}
+
+
+async def reconcile_recoveries(service: OperationalTaskService, timescale: AsyncConnection,
+                               settings: ProjectSettings, *, limit: int = 20,
+                               created_from: datetime | None = None) -> int:
+    """Poll only persisted recovery waits; publish evidence before waking the Agent."""
+    from app.domain.schemas.operational_task import OperationalTaskSummary, StoredEvidenceRecord
+    from app.domain.schemas.operational_data import TaskDataSnapshot
+    from app.infra.db.timescaledb.repositories.scada import ScadaRepository
+
+    async with service.repo.transaction():
+        candidates = await service.repo.recovery_candidates(limit, created_from)
+    resumed = 0
+    for task in candidates:
+        context = task["handling_context"]
+        returned_at = datetime.fromisoformat(context["returned_at"])
+        incident_at = datetime.fromisoformat(task["trigger"]["occurred_at"])
+        interval = settings.scada_monitoring_interval_seconds
+        roles = settings.scada_monitoring
+        devices = list(dict.fromkeys(roles.pressure_device_ids + roles.outlet_flow_device_ids)) if roles else []
+        source = "scada_readings"
+        through = min(service._now(), returned_at + timedelta(days=1))
+        if context.get("device_ids"):
+            devices = context["device_ids"]
+            interval = context.get("interval_seconds", interval)
+        else:
+            # Retain the points the Agent actually inspected, including water
+            # quality or local incident sensors outside the project role lists.
+            for evidence in await service.repo.list_evidence(task["task_id"]):
+                saved = evidence.get("data_snapshot") or {}
+                if saved.get("source") not in {"query_scada_readings", "scada_readings"}:
+                    continue
+                data = saved.get("data")
+                if isinstance(data, dict):
+                    data = data.get("data", data.get("items", []))
+                if isinstance(data, list):
+                    devices.extend(row["device_id"] for row in data if isinstance(row, dict) and isinstance(row.get("device_id"), str))
+            devices = list(dict.fromkeys(devices))
+        async with timescale.transaction():
+            rows = await ScadaRepository.get_scada_by_ids_time_range(
+                timescale, devices, incident_at - timedelta(days=15), through, row_limit=100_001,
+            ) if through > returned_at and devices else []
+        result = ({"status": "waiting", "reason": "reading_budget_exceeded"} if len(rows) > 100_000 else
+                  evaluate_recovery(rows, device_ids=devices, returned_at=returned_at,
+                                    incident_at=incident_at, interval_seconds=interval or 0))
+        async with service.repo.transaction():
+            await service.repo.lock(f"task:{task['task_id']}")
+            current = await service._task(UUID(task["task_id"]))
+            # A new receipt, cancellation or another poll may have won the race.
+            if current.get("handling_context") != context or current["status"] in {"cancelled", "failed"}:
+                continue
+            now = service._now()
+            if devices and interval:
+                current["handling_context"].update(device_ids=devices, interval_seconds=interval)
+            current["handling_context"]["last_checked_at"] = now.isoformat()
+            current["handling_context"]["last_check"] = result
+            if result["status"] == "passed":
+                evidence_id = "recovery:" + digest({"task_id": task["task_id"], "returned_at": context["returned_at"], "result": result})
+                evidence = StoredEvidenceRecord(
+                    evidence_id=evidence_id, kind="system_data",
+                    summary="完工反馈后连续两个采样周期的 SCADA 恢复正常，等待 Agent 完成复核报告",
+                    source_label="恢复供水监测复核", source_ref=f"operational-task:{task['task_id']}",
+                    observed_at=datetime.fromisoformat(result["verified_through"]), recorded_at=now,
+                    recorded_by=service._actor(), has_snapshot=True,
+                    data_snapshot=TaskDataSnapshot(source=source, source_ref=evidence_id, queried_at=now,
+                                                   origin="system_data", data=result, truncated=False),
+                )
+                if not await service.repo.get_evidence(task["task_id"], evidence_id):
+                    await service.repo.insert_evidence(task["task_id"], evidence.model_dump(mode="json"))
+                current["handling_context"].update(phase="verification_ready", verification_evidence_id=evidence_id)
+                await service._requeue_task(current, now)
+                resumed += 1
+            else:
+                current.update(revision=current["revision"] + 1, updated_at=now)
+            await service.repo.update_task(OperationalTaskSummary.model_validate(current).model_dump(
+                mode="json", exclude={"active_run", "latest_run"}))
+            await service._task_event(current, now)
+    return resumed

app/services/operational_tasks.py

@@ -72,6 +72,7 @@ from app.domain.schemas.operational_task import (
 from app.domain.schemas.operational_data import TaskDataSnapshot, TaskDataSource
 from app.services.content_digest import digest
 from app.services.operational_presentation import resolve_overlay
+from app.services.operational_recovery import accept_completion, track_work_order
 
 
 LEASE_DURATION = timedelta(seconds=60)
@@ -1005,7 +1006,7 @@ class OperationalTaskService:
                 # Compatibility for pre-015 repository adapters only.
                 if command.report:
                     if task["task_type"] == "event_investigation":
-                        task["status"] = self._investigation_status(actions.values()) if command.report.stage == "final" else "running"
+                        task["status"] = self._investigation_status(actions.values(), task) if command.report.stage == "final" else "running"
                     else:
                         task["status"] = self._status_from_plan(current_plan.plan.nodes, actions.values()) if current_plan else "running"
                     if task["status"] in {"waiting_human", "completed", "failed", "cancelled"}:
@@ -1069,8 +1070,6 @@ class OperationalTaskService:
             await self.repo.lock(f"run:{command.run_id}")
             task = await self._task(task_id)
             self._require_actionable_task(task)
-            require(not task.get("source", {}).get("replay_context"), "REPLAY_ACTION_DISABLED",
-                    "SCADA 回放调查仅生成分析与报告，不执行真实工单或设备动作", status=409)
             run = await self._active_run(task)
             require(run is not None and str(run["run_id"]) == str(command.run_id)
                     and run["generation"] == task["execution_generation"],
@@ -1133,6 +1132,7 @@ class OperationalTaskService:
             await self.repo.lock(f"task:{task_id}")
             task = await self._task(task_id)
             self._require_actionable_task(task)
+            self._require_real_action(task)
             self._validate_action_payload(command)
             if task["task_type"] == "event_investigation":
                 report = await self.repo.get_latest_report(str(task_id))
@@ -1201,7 +1201,9 @@ class OperationalTaskService:
                 return HumanApprovalAction.model_validate(replay["action"])
             await self.repo.lock(f"task:{task_id}")
             await self.repo.lock(f"action:{action_id}")
-            self._require_actionable_task(await self._task(task_id))
+            task = await self._task(task_id)
+            self._require_actionable_task(task)
+            self._require_real_action(task)
             action = await self._action(task_id, action_id)
             require(action.kind == "create_work_order", "INVALID_ACTION_KIND",
                     "Only work-order drafts can be edited", status=422)
@@ -1255,7 +1257,9 @@ class OperationalTaskService:
                 return HumanApprovalAction.model_validate(replay["action"])
             await self.repo.lock(f"task:{task_id}")
             await self.repo.lock(f"action:{action_id}")
-            self._require_actionable_task(await self._task(task_id))
+            task = await self._task(task_id)
+            self._require_actionable_task(task)
+            self._require_real_action(task)
             action = await self._action(task_id, action_id)
             require(action.version == command.expected_version, "VERSION_CONFLICT",
                     "Action version changed", status=409)
@@ -1300,7 +1304,9 @@ class OperationalTaskService:
                 return HumanApprovalAction.model_validate(replay["action"])
             await self.repo.lock(f"task:{task_id}")
             await self.repo.lock(f"action:{action_id}")
-            self._require_actionable_task(await self._task(task_id))
+            task = await self._task(task_id)
+            self._require_actionable_task(task)
+            self._require_real_action(task)
             action = await self._action(task_id, action_id)
             require(action.version == command.expected_version, "VERSION_CONFLICT",
                     "Action version changed", status=409)
@@ -1396,6 +1402,7 @@ class OperationalTaskService:
                 )
             if outcome.status in {"succeeded", "failed", "unsupported"}:
                 task_document = await self._task(task_id)
+                track_work_order(task_document, _json(current))
                 await self._requeue_task(task_document, outcome_recorded_at)
                 await self.repo.update_task(
                     _task_json(OperationalTaskSummary.model_validate(task_document))
@@ -1469,6 +1476,7 @@ class OperationalTaskService:
                 )
             if outcome.status in {"succeeded", "failed", "unsupported"}:
                 task_document = await self._task(task_id)
+                track_work_order(task_document, _json(current))
                 await self._requeue_task(task_document, outcome_recorded_at)
                 await self.repo.update_task(
                     _task_json(OperationalTaskSummary.model_validate(task_document))
@@ -1589,6 +1597,10 @@ class OperationalTaskService:
                         ),
                     )
             if task["task_type"] == "event_investigation":
+                # Recover older tasks whose successful dispatch predates tracking.
+                for item in await self.repo.list_actions(str(task_id)):
+                    track_work_order(task, item)
+                accept_completion(task, body)
                 if matching_source:
                     task["source"] = {**task["source"], "external_revision": command.source_revision}
                 await self._requeue_task(task, now)
@@ -2032,7 +2044,15 @@ class OperationalTaskService:
                         for item in await self._list_access_requests(task_id)
                     ]
                     if task["task_type"] == "event_investigation":
-                        lifecycle = self._investigation_status(actions)
+                        lifecycle = self._investigation_status(actions, task)
+                        recovery = task.get("handling_context") or {}
+                        if lifecycle == "completed" and recovery:
+                            require(recovery.get("verification_evidence_id") in report.report.evidence_refs,
+                                    "RECOVERY_EVIDENCE_REQUIRED", "最终复核报告必须引用恢复监测证据", status=409)
+                            require(report.report.outcome == "concluded", "RECOVERY_EVIDENCE_REQUIRED",
+                                    "恢复复核通过后才能结束事件", status=409)
+                            recovery.update(phase="completed", title="恢复供水复核成功", completed_at=now.isoformat())
+                            await self._resolve_investigation_alerts(str(task_id), recovery["verification_evidence_id"])
                     elif report.report.outcome == "insufficient_data":
                         terminal = all(node.status in {"completed", "skipped", "cancelled"}
                                        for node in plan.plan.nodes)
@@ -2081,7 +2101,8 @@ class OperationalTaskService:
                     failure_detail or "工况任务执行失败",
                     now,
                 )
-            elif lifecycle == "waiting_human" and not pending_question:
+            elif (lifecycle == "waiting_human" and not pending_question
+                  and (task.get("handling_context") or {}).get("phase") != "waiting_scada"):
                 pending_actions = any(
                     action.approval_status == "pending" for action in actions
                 )
@@ -2108,11 +2129,32 @@ class OperationalTaskService:
         )
         return run
 
+    async def _resolve_investigation_alerts(self, task_id: str, evidence_id: str) -> None:
+        from app.domain.schemas.task_alert import TaskAlertEvent
+        from app.infra.db.postgresql.task_alerts import TaskAlertRepository
+        from app.services.task_alerts import TaskAlertService
+
+        if not hasattr(self.repo, "list_investigation_alerts"):
+            return
+        for alert in await self.repo.list_investigation_alerts(task_id):
+            command = TaskAlertEvent.model_validate({
+                **{key: value for key, value in alert.items() if key not in {"status", "updated_at", "revision"}},
+                "event_id": uuid5(NAMESPACE_URL, f"recovery:{task_id}:{evidence_id}:{alert['alert_id']}"),
+                "event_type": "resolved",
+            })
+            await TaskAlertService(TaskAlertRepository(self.repo.connection, self.repo.project_id), None,
+                                   internal_producer_id=command.producer_id).push(command)
+
     async def _task(self, task_id: UUID) -> dict:
         task = await self.repo.get_task(str(task_id))
         require(task is not None, "TASK_NOT_FOUND", "Operational task not found", status=404)
         return task
 
+    @staticmethod
+    def _require_real_action(task: dict) -> None:
+        require(not (task.get("source") or {}).get("replay_context"), "REPLAY_ACTION_DISABLED",
+                "SCADA 回放调查仅生成分析与报告，不执行真实工单或设备动作", status=409)
+
     @staticmethod
     def _require_actionable_task(task: dict) -> None:
         source = task.get("source") or {}
@@ -2176,8 +2218,10 @@ class OperationalTaskService:
         return "running"
 
     @staticmethod
-    def _investigation_status(actions: Iterable[HumanApprovalAction]) -> str:
-        """Only recorded business actions can keep an investigation waiting."""
+    def _investigation_status(actions: Iterable[HumanApprovalAction], task: dict | None = None) -> str:
+        """Field completion and recovery evidence precede incident closure."""
+        if (task or {}).get("handling_context", {}).get("phase") in {"waiting_work_order", "waiting_scada"}:
+            return "waiting_human"
         return "waiting_human" if any(
             action.approval_status == "pending" or (
                 action.approval_status == "approved"

app/services/scada_monitoring_calibration.py

@@ -0,0 +1,141 @@
+"""Fit causal replay thresholds in bounded numeric arrays, before publishing history."""
+from datetime import datetime, timedelta
+
+import numpy as np
+import pandas as pd
+
+from app.domain.schemas.project_settings import ScadaMonitoringSettings
+from app.domain.schemas.scada_replay import ScadaReplayCreate
+from app.services.scada_normal_profile import CALIBRATION_VERSION, DIAGNOSTIC_POLICY
+
+
+def adaptive_score_threshold(scores: list[float]) -> dict:
+    """Keep the cutoff below the normal lower tail, with a fixed noise margin."""
+    values = np.asarray(scores, dtype=float)
+    if len(values) < 5 or not np.isfinite(values).all():
+        return {"status": "insufficient_data", "reason": "normal_score_history_incomplete"}
+    center = float(np.median(values))
+    sigma = float(1.4826 * np.median(np.abs(values - center)))
+    lower_tail = float(np.quantile(values, 0.05))
+    return {
+        "status": "ready",
+        "threshold": min(-0.04, center - 3 * sigma, lower_tail - 0.02),
+        "normal_score_median": center,
+        "normal_score_min": float(values.min()),
+        "normal_score_max": float(values.max()),
+        "normal_score_robust_sigma": sigma,
+        "normal_score_lower_quantile": lower_tail,
+        "normal_score_count": len(values),
+        "method": "min(-0.04, median-3*robust_sigma, normal_q05-0.02)",
+    }
+
+
+class MonitoringCalibrationBuilder:
+    """Retain bounded numeric series, never full raw pages or device batches."""
+
+    def __init__(self, document: dict, config: ScadaReplayCreate):
+        self.settings = ScadaMonitoringSettings.model_validate(document["monitoring_config"])
+        self.config = config
+        self.cutoff = config.start_time
+        self.start = self.cutoff - timedelta(days=15)
+        self.interval = int(document["monitoring_interval_seconds"])
+        self.supported = self.interval % 60 == 0 and 86400 % self.interval == 0
+        self.count = 15 * (86400 // self.interval) if self.supported else 0
+        ids = dict.fromkeys(self.settings.pressure_device_ids + self.settings.outlet_flow_device_ids
+                            + [device for rule in config.rules if rule.kind == "relative_change"
+                               for device in rule.device_ids])
+        self.values = {device: np.full(self.count, np.nan) for device in ids}
+
+    def add(self, rows: list[dict]) -> None:
+        for row in rows:
+            device = row["device_id"]
+            if device not in self.values:
+                continue
+            at = datetime.fromisoformat(row["time"])
+            if at >= self.cutoff:
+                raise ValueError("动态阈值拟合不能读取回放开始时刻及之后的观测")
+            offset = (at - self.start).total_seconds()
+            if (not self.supported or offset < 0 or offset % self.interval
+                    or row.get("quality") not in {"valid", "corrected"}
+                    or row.get("cleaned_value") is None):
+                continue
+            index = int(offset // self.interval)
+            if np.isfinite(self.values[device][index]):
+                raise ValueError("动态阈值拟合存在重复的设备时间点")
+            self.values[device][index] = float(row["cleaned_value"])
+
+    def fit(self) -> dict:
+        base = {
+            "version": CALIBRATION_VERSION,
+            "diagnostic_policy": DIAGNOSTIC_POLICY,
+            "knowledge_cutoff": self.cutoff.isoformat(),
+            "history_start": self.start.isoformat(),
+            "required_samples_per_device": self.count,
+            "pressure": {}, "outlet_flow": {}, "diagnostic": {},
+        }
+        if not self.supported:
+            return {**base, "status": "insufficient_data", "reason": "unsupported_sampling_interval"}
+        available = {device: int(np.isfinite(values).sum()) for device, values in self.values.items()}
+        # A regular diurnal rise/reversal can be a huge percentage of a small
+        # preceding flow. Judge its absolute value against causal normal readings
+        # at the same sampling phase before treating that percentage as abnormal.
+        points_per_day = 86400 // self.interval
+        for device, values in self.values.items():
+            if available[device] != self.count:
+                continue
+            days = values.reshape(15, points_per_day)
+            center = np.median(days, axis=0)
+            sigma = 1.4826 * np.median(np.abs(days - center), axis=0)
+            margin = np.maximum(3 * sigma, np.maximum(np.abs(center) * 1e-6, 1e-9))
+            base["diagnostic"][device] = {
+                "phase_origin": self.start.isoformat(), "interval_seconds": self.interval,
+                "lower": (days.min(axis=0) - margin).tolist(),
+                "upper": (days.max(axis=0) + margin).tolist(),
+                "method": "normal_same_phase_range_plus_3_robust_sigma", "normal_day_count": 15,
+            }
+        for kind, ids, direction, floor in (
+            ("pressure", self.settings.pressure_device_ids, -1, 3.0),
+            ("outlet_flow", self.settings.outlet_flow_device_ids, 1, 5.0),
+        ):
+            for device in ids:
+                if available[device] != self.count:
+                    continue
+                rule = next((rule for rule in self.config.rules
+                             if rule.kind == "relative_change" and device in rule.device_ids), None)
+                if rule is None:
+                    continue
+                series = pd.Series(self.values[device])
+                reference = series.rolling(rule.window_samples, min_periods=rule.minimum_samples).median().shift(1)
+                changes = (direction * 100 * (series - reference) / reference.abs()).to_numpy()
+                changes = changes[np.isfinite(changes)]
+                if not len(changes):
+                    continue
+                center = float(np.median(changes))
+                sigma = float(1.4826 * np.median(np.abs(changes - center)))
+                base[kind][device] = {
+                    "threshold_percent": max(floor, center + 3 * sigma),
+                    "normal_change_median": center,
+                    "normal_change_robust_sigma": sigma,
+                    "normal_change_count": int(len(changes)),
+                    "method": "normal_recent_window_change_median_plus_3_robust_sigma",
+                    "normal_profile": base["diagnostic"].get(device),
+                }
+        pressure_ids = self.settings.pressure_device_ids
+        incomplete = {device: available[device] for device in pressure_ids if available[device] != self.count}
+        if len(pressure_ids) < 5 or incomplete:
+            return {**base, "status": "insufficient_data", "reason": "pressure_history_incomplete",
+                    "available_samples": available}
+        from app.services import scada_replay_monitoring as monitoring
+
+        index = pd.date_range(start=self.start, periods=self.count, freq=pd.Timedelta(seconds=self.interval))
+        frame = pd.DataFrame({device: self.values[device] for device in pressure_ids}, index=index)
+        result = monitoring.run_burst_detection(
+            network="scada-replay-baseline", username="scada-replay",
+            observed_pressure_data=frame, points_per_day=86400 // self.interval,
+            mu=monitoring.TARGET_MU,
+            iforest_params={"n_estimators": monitoring.TARGET_N_ESTIMATORS,
+                            "random_state": monitoring.TARGET_RANDOM_STATE, "contamination": "auto"},
+            sensor_nodes=pressure_ids,
+        )
+        score = adaptive_score_threshold([float(row["Score"]) for row in result["rows"]])
+        return {**base, "status": score["status"], "burst_detection": score}

app/services/scada_monitoring_incidents.py

@@ -0,0 +1,98 @@
+"""Persist one accident across monitoring periods without moving its evidence cutoff."""
+from datetime import datetime
+from math import isfinite
+from uuid import NAMESPACE_URL, uuid5
+
+
+def _anchors(monitoring: dict, indicator: str = "pressure") -> dict:
+    return {
+        item["device_id"]: {"baseline": item["baseline"], "threshold_percent": item["threshold_percent"]}
+        for item in monitoring.get(indicator, {}).get("devices", [])
+        if item.get("hit") and item.get("baseline", 0) > 0
+        and 0 < item.get("threshold_percent", 0) < 100
+    }
+
+
+def update_incident(document: dict, monitoring: dict, sequence: int) -> tuple[dict | None, str | None]:
+    incident = document.get("monitoring_incident")
+    # Adopt the last task from an older replay document on its first new step.
+    previous = document.get("last_monitoring") or {}
+    if incident:
+        incident.setdefault("burst_diagnosed", previous.get("status") == "suspected_burst")
+    if not incident and previous.get("task_id"):
+        incident = {
+            "incident_id": str(uuid5(NAMESPACE_URL, f"scada_replay:{document['replay_id']}:monitor:"
+                                    f"{previous['period_end']}:{previous['rule_version']}")),
+            "started_at": previous["period_end"], "last_observed_at": previous["period_end"],
+            "status": "active", "task_id": previous["task_id"], "recovery_count": 0,
+            "pressure_anchors": _anchors(previous), "alert": None,
+            "flow_anchors": _anchors(previous, "outlet_flow"),
+            "burst_diagnosed": previous.get("status") == "suspected_burst",
+        }
+        document["monitoring_incident"] = incident
+    if incident and incident.get("last_monitoring_id") == monitoring.get("monitoring_id") and monitoring.get("monitoring_id"):
+        return incident, incident["last_action"]
+    anomaly = (monitoring["status"] in {"attention", "suspected_burst"}
+               or bool(monitoring.get("anomaly_signals"))
+               or any(monitoring.get(key, {}).get("hit") is True
+                      for key in ("pressure", "outlet_flow", "burst_detection")))
+    if anomaly and (not incident or incident["status"] == "resolved"):
+        external = (f"scada_replay:{document['replay_id']}:monitor:"
+                    f"{monitoring['period_end']}:{monitoring['rule_version']}")
+        incident = {
+            "incident_id": str(uuid5(NAMESPACE_URL, external)), "external_id": external,
+            "started_at": monitoring["period_end"], "status": "active", "task_id": None,
+            "recovery_count": 0, "pressure_anchors": _anchors(monitoring), "alert": None,
+            "flow_anchors": _anchors(monitoring, "outlet_flow"),
+            "batch_sequence": sequence,
+        }
+        document["monitoring_incident"] = incident
+        action = "raised"
+    elif incident and incident["status"] == "active":
+        # A later diagnostic stage may add pressure/flow evidence to the same task.
+        # Freeze each newly abnormal device once; retain earlier references.
+        for indicator, field in (("pressure", "pressure_anchors"), ("outlet_flow", "flow_anchors")):
+            for device, anchor in _anchors(monitoring, indicator).items():
+                incident.setdefault(field, {}).setdefault(device, anchor)
+        current = {item["device_id"]: item for item in monitoring.get("pressure", {}).get("devices", [])}
+        anchors = incident["pressure_anchors"]
+        flow = {item["device_id"]: item for item in monitoring.get("outlet_flow", {}).get("devices", [])}
+        # Rolling relative changes can become normal while absolute pressure stays low.
+        # Confirm recovery against the frozen pre-accident references as well.
+        recovered = monitoring["status"] == "normal" and not anomaly and all(
+            monitoring.get(key, {}).get("status") == "ready"
+            and monitoring[key].get("hit") is False for key in ("pressure", "burst_detection")
+        ) and all(
+            device in current and isfinite(current[device]["value"])
+            and current[device].get("observed_at") is not None
+            and datetime.fromisoformat(current[device]["observed_at"]) > datetime.fromisoformat(incident["last_observed_at"])
+            and current[device]["value"] >= anchor["baseline"] * (1 - anchor["threshold_percent"] / 100)
+            for device, anchor in anchors.items()
+        ) and all(
+            device in flow and isfinite(flow[device]["value"])
+            and flow[device].get("observed_at") is not None
+            and datetime.fromisoformat(flow[device]["observed_at"]) > datetime.fromisoformat(incident["last_observed_at"])
+            and flow[device]["value"] <= anchor["baseline"] * (1 + anchor["threshold_percent"] / 100)
+            for device, anchor in incident.get("flow_anchors", {}).items()
+        )
+        contiguous = bool(monitoring.get("period_start")) and (
+            datetime.fromisoformat(monitoring["period_start"]) == datetime.fromisoformat(incident["last_observed_at"])
+        )
+        incident["recovery_count"] = (incident["recovery_count"] + 1 if contiguous else 1) if recovered else 0
+        action = "resolved" if incident["recovery_count"] >= 2 else "ongoing"
+        if action == "resolved":
+            incident["status"] = "resolved"
+    else:
+        return None, None
+    incident.update(last_observed_at=monitoring["period_end"],
+                    last_monitoring_id=monitoring.get("monitoring_id"), last_action=action)
+    return incident, action
+
+
+def attach_incident(monitoring: dict, incident: dict, action: str) -> None:
+    monitoring["task_id"] = incident["task_id"]
+    monitoring["incident"] = {
+        key: incident[key] for key in (
+            "incident_id", "status", "started_at", "last_observed_at", "recovery_count",
+        )
+    } | {"action": action}

app/services/scada_normal_profile.py

@@ -0,0 +1,48 @@
+"""Clock-phase normal envelopes frozen before replay; never inspect future readings."""
+from datetime import datetime
+from math import isfinite
+
+CALIBRATION_VERSION = "scada-monitor-calibration@2"
+DIAGNOSTIC_POLICY = "normal-phase-envelope@1"
+
+
+def weighted_diagnosis(document: dict) -> bool:
+    return (document.get("monitoring_config") or {}).get("decision_policy") == "weighted"
+
+
+def current_calibration(document: dict, start: datetime) -> dict | None:
+    calibration = document.get("monitoring_calibration") or {}
+    if (calibration.get("version") == CALIBRATION_VERSION
+            and calibration.get("diagnostic_policy") == DIAGNOSTIC_POLICY
+            and calibration.get("knowledge_cutoff") == start.isoformat()):
+        return calibration
+    return None
+
+
+def diagnostic_signal(document: dict, event: dict) -> bool:
+    """Never revive a persisted fixed-percentage event as new diagnosis evidence."""
+    if event.get("category") == "data_quality":
+        return False
+    if not weighted_diagnosis(document):
+        return True
+    rules = (document.get("config") or {}).get("rules", [])
+    rule = next((item for item in rules if item["rule_id"] == event["rule_id"]), None)
+    return bool(rule and (rule["kind"] == "bounds" or event.get("diagnostic_policy") == DIAGNOSTIC_POLICY))
+
+
+def normal_profile_bounds(profile: dict | None, at: datetime) -> tuple[float, float] | None:
+    if not profile:
+        return None
+    interval = profile.get("interval_seconds", 0)
+    if not profile.get("phase_origin"):
+        return None
+    origin = datetime.fromisoformat(profile["phase_origin"])
+    lower, upper = profile.get("lower", []), profile.get("upper", [])
+    offset = (at - origin).total_seconds()
+    if interval <= 0 or offset < 0 or offset % interval or not lower or len(lower) != len(upper):
+        return None
+    slot = int(offset // interval) % len(lower)
+    low, high = lower[slot], upper[slot]
+    if low is None or high is None or not isfinite(low) or not isfinite(high):
+        return None
+    return float(low), float(high)

app/services/scada_replay.py

@@ -2,6 +2,7 @@
 from __future__ import annotations
 
 from datetime import datetime, timedelta, timezone
+import json
 from uuid import NAMESPACE_URL, uuid4, uuid5
 
 from fastapi import HTTPException
@@ -14,6 +15,18 @@ from app.infra.db.timescaledb.repositories.scada import ScadaRepository
 from app.services.content_digest import digest
 from app.services.scada_replay_processing import prepare_cleaning, detect_frame
 from app.services.scada_replay_monitoring import evaluate_monitoring_period
+from app.services.scada_normal_profile import current_calibration, diagnostic_signal, weighted_diagnosis
+
+
+# 500 devices at 15-minute sampling need 720,000 rows for fifteen days.
+# Leave headroom without allowing dense sources to make preparation unbounded.
+BASELINE_PAGE_SIZE = 2000
+BASELINE_MAX_ROWS = 1_000_000
+BASELINE_MAX_BYTES = 512 * 1024 * 1024
+
+
+class BaselineBudgetExceeded(RuntimeError):
+    """The unpublished baseline exceeded its fixed preparation budget."""
 
 
 def parse_time(value: str) -> datetime:
@@ -36,6 +49,10 @@ def snapshot(document: dict) -> ScadaReplaySnapshot:
         step_pending=document.get("step_pending", False),
         monitoring_count=document.get("monitoring_count", 0),
         last_monitoring=document.get("last_monitoring"),
+        # Old persisted documents without a policy still use three-signal detection.
+        monitoring_config=({"decision_policy": "all_signals", **document["monitoring_config"]}
+                           if document.get("monitoring_config") else None),
+        monitoring_calibration=document.get("monitoring_calibration"),
     )
 
 
@@ -105,7 +122,7 @@ class ScadaReplayService:
         await self.ensure_storage()
         async with self.repo.control():
             document = await self.repo.get(replay_id)
-            key, body = str(command.request_id), command.model_dump(mode="json")
+            key, body = str(command.request_id), command.model_dump(mode="json", exclude_none=True)
             receipt = document.setdefault("receipts", {}).get(key)
             if receipt:
                 if receipt != body:
@@ -114,9 +131,13 @@ class ScadaReplayService:
             if command.expected_revision != document["revision"]:
                 raise HTTPException(409, "回放控制状态已更新，请刷新后重试")
             action, status = command.action, document["status"]
-            if document.get("step_pending") and action == "step":
+            if (action in {"step", "monitor", "retry"} and document.get("warmed")
+                    and weighted_diagnosis(document)
+                    and current_calibration(document, parse_time(document["start_time"])) is None):
+                raise HTTPException(409, "旧回放缺少新版历史诊断校准，请停止后新建演示")
+            if document.get("step_pending") and action in {"step", "monitor"}:
                 raise HTTPException(409, "当前时间步正在处理，请完成后继续")
-            if action in {"step", "retry"} and await self.repo.active_other(replay_id):
+            if action in {"step", "monitor", "retry"} and await self.repo.active_other(replay_id):
                 raise HTTPException(409, "当前项目已有另一条活动回放")
             if action == "step" and status in {"ready", "paused"}:
                 interval = document.get("monitoring_interval_seconds")
@@ -126,8 +147,31 @@ class ScadaReplayService:
                 elapsed = (parse_time(document["virtual_now"]) - start).total_seconds()
                 target = start + timedelta(seconds=(int(elapsed // interval) + 1) * interval)
                 document["step_start"] = document["virtual_now"]
+                document["step_anomaly_signals"] = []
                 document["step_target"] = min(parse_time(document["end_time"]), target).isoformat()
                 document["status"], document["step_pending"] = "paused", True
+            elif action == "monitor" and status in {"ready", "paused", "completed"}:
+                target = command.at_time
+                if target is None:
+                    raise HTTPException(422, "选定时刻监测需要提供 at_time")
+                start = parse_time(document["start_time"])
+                current = parse_time(document["virtual_now"])
+                if not start <= target <= parse_time(document["end_time"]):
+                    raise HTTPException(422, "监测时刻必须在本轮演示区间内")
+                if target < current:
+                    raise HTTPException(409, "监测历史时刻需要新建一轮回放，避免使用后续状态")
+                interval = document.get("monitoring_interval_seconds")
+                if interval is None:
+                    raise HTTPException(409, "本轮未固定监测周期，请停止后新建演示")
+                previous = document.get("last_monitoring") or {}
+                # Reusing an immutable result must not count as another normal
+                # period, resolve an incident, or create another investigation.
+                if not previous.get("period_end") or parse_time(previous["period_end"]) != target:
+                    document["step_start"] = (current if target > current
+                        else max(start, target - timedelta(seconds=interval))).isoformat()
+                    document["step_target"] = target.isoformat()
+                    document["step_anomaly_signals"] = []
+                    document["status"], document["step_pending"] = "paused", True
             elif action == "stop" and status not in {"completed", "stopped"}:
                 document["status"], document["step_pending"] = "stopped", False
                 document.pop("step_target", None)
@@ -161,11 +205,16 @@ class ScadaReplayService:
                     # Failure never discards the prepared batch; retry resumes it.
                     document = await self.repo.get(replay_id)
                     document["status"], document["step_pending"] = "failed", False
-                    document["error"] = (
-                        "SOURCE_CHANGED: 原始观测已改变，需要新建回放"
-                        if "SOURCE_CHANGED" in str(exc)
-                        else "SCADA 处理失败，已保留批次与进度；请检查服务日志后重试"
-                    )
+                    if isinstance(exc, HTTPException):
+                        document["error"] = str(exc.detail)
+                    elif isinstance(exc, BaselineBudgetExceeded):
+                        document["error"] = str(exc)
+                    else:
+                        document["error"] = (
+                            "SOURCE_CHANGED: 原始观测已改变，需要新建回放"
+                            if "SOURCE_CHANGED" in str(exc)
+                            else "SCADA 处理失败，已保留批次与进度；请检查服务日志后重试"
+                        )
                     async with self.repo.connection.transaction():
                         await self.repo.save(document)
                     raise
@@ -174,23 +223,9 @@ class ScadaReplayService:
         config = ScadaReplayCreate.model_validate(document["config"])
         replay_id = document["replay_id"]
         if not document.get("warmed"):
-            import math
-            window = max(rule.window_samples for rule in config.rules)
-            async with self.timescale.transaction():
-                warmup = await ScadaRepository.get_warmup_values(
-                    self.timescale, config.device_ids, config.start_time, window,
-                )
-            for row in warmup:
-                value = row["monitored_value"]
-                bounds = config.quality_ranges.get(row["device_id"])
-                if value is not None and math.isfinite(value) and not (bounds and (
-                    bounds.lower is not None and value < bounds.lower
-                    or bounds.upper is not None and value > bounds.upper
-                )):
-                    document["histories"].setdefault(row["device_id"], []).append(value)
-            document["warmed"] = True
-            async with self.repo.connection.transaction():
-                await self.repo.save(document)
+            await self._freeze_baseline(document, config)
+        if weighted_diagnosis(document) and current_calibration(document, config.start_time) is None:
+            raise HTTPException(409, "旧回放缺少新版历史诊断校准，请停止后新建演示")
         batch = await self.repo.pending_batch(replay_id)
         if batch is None:
             step = document.get("step_pending", False)
@@ -252,6 +287,15 @@ class ScadaReplayService:
             target = result.get("step_target")
             was_step = bool(target)
             result["step_pending"] = bool(target and parse_time(batch["at"]) < parse_time(target))
+            if was_step:
+                signals = {item["event_id"]: item for item in result.get("step_anomaly_signals", [])
+                           if diagnostic_signal(result, item)}
+                for event in changes:
+                    if event["change"] == "raised" and diagnostic_signal(result, event):
+                        signals[event["event_id"]] = {key: event[key] for key in (
+                            "event_id", "rule_id", "device_ids", "detail", "diagnostic_policy"
+                        ) if key in event}
+                result["step_anomaly_signals"] = list(signals.values())
             monitoring = None
             if was_step and not result["step_pending"]:
                 period_start = parse_time(result.get("step_start", result["start_time"]))
@@ -261,7 +305,7 @@ class ScadaReplayService:
                     role.get("pressure_device_ids", []) + role.get("outlet_flow_device_ids", [])
                 ))
                 previous = await self.repo.monitoring_outputs(
-                    replay_id, role_ids, max(config.start_time, period_end - timedelta(days=15)),
+                    replay_id, role_ids, period_end - timedelta(days=15),
                     period_end, batch["sequence"],
                 ) if role_ids else []
                 from app.mcp.analysis_service import bounded_calculation
@@ -273,11 +317,25 @@ class ScadaReplayService:
                     period_end=period_end,
                     batch_sequence=batch["sequence"],
                 )
-                result["last_monitoring"] = monitoring
+                # Any configured measurement rule can enter diagnosis, including
+                # indicators outside the pressure/outlet roles used for burst scoring.
+                signals = {item["event_id"]: item for item in result.get("step_anomaly_signals", [])
+                           if diagnostic_signal(result, item)}
+                for state in result.get("rule_states", {}).values():
+                    event = state.get("event")
+                    if event and diagnostic_signal(result, event):
+                        signals[event["event_id"]] = {key: event[key] for key in (
+                            "event_id", "rule_id", "device_ids", "detail", "diagnostic_policy"
+                        ) if key in event}
+                monitoring["anomaly_signals"] = list(signals.values())
+                if signals and monitoring["status"] == "normal":
+                    monitoring["status"] = "attention"
+                    monitoring["detail"] = "本周期曾检出测点异常，进入异常调查与爆管诊断"
                 result["monitoring_count"] = result.get("monitoring_count", 0) + 1
             if not result["step_pending"]:
                 result.pop("step_target", None)
                 result.pop("step_start", None)
+                result.pop("step_anomaly_signals", None)
             if batch["refs"]:
                 last = batch["refs"][-1]
                 result["cursor"] = [last["time"], last["device_id"]]
@@ -295,6 +353,80 @@ class ScadaReplayService:
                 await self.repo.save_batch(replay_id, batch)
                 await self.repo.save(result)
 
+    async def _freeze_baseline(self, document, config):
+        """Stage bounded pages, then atomically publish one causal source snapshot."""
+        if document.get("warmed"):
+            return
+        start = config.start_time - timedelta(days=15)
+        batch_count, row_count, byte_count, after = 0, 0, 0, None
+        counts = {device_id: 0 for device_id in config.device_ids}
+        valid_counts = dict(counts)
+        histories = {device_id: [] for device_id in config.device_ids}
+        window = max(rule.window_samples for rule in config.rules)
+        calibration = None
+        if (document.get("monitoring_config") or {}).get("decision_policy") == "weighted":
+            from app.services.scada_monitoring_calibration import MonitoringCalibrationBuilder
+            calibration = MonitoringCalibrationBuilder(document, config)
+        async with self.repo.connection.transaction():
+            await self.repo.clear_unpublished_baseline(document["replay_id"])
+        # Keep only the current page in memory. Each BizDB write commits before
+        # the next source read; clean_prepared baseline pages remain invisible.
+        async with self.timescale.transaction():
+            await ScadaRepository.begin_replay_baseline_snapshot(self.timescale)
+            while True:
+                rows = await ScadaRepository.get_raw_page(
+                    self.timescale, config.device_ids, start, config.start_time,
+                    config.start_time, min(BASELINE_PAGE_SIZE, BASELINE_MAX_ROWS - row_count + 1), after,
+                )
+                if not rows:
+                    break
+                if row_count + len(rows) > BASELINE_MAX_ROWS:
+                    raise BaselineBudgetExceeded(
+                        f"BASELINE_BUDGET_EXCEEDED: 十五天历史基线超过 {BASELINE_MAX_ROWS:,} 条准备预算，"
+                        "尚未发布；请减少回放设备或核查历史采样密度后重试"
+                    )
+                outputs = prepare_cleaning(rows, config)
+                if calibration is not None:
+                    calibration.add(outputs)
+                batch = {
+                    "sequence": -batch_count - 1, "stage": "clean_prepared", "kind": "baseline",
+                    "at": outputs[-1]["time"], "outputs": outputs, "events": [],
+                }
+                batch_bytes = len(json.dumps(batch, ensure_ascii=False, allow_nan=False,
+                                             separators=(",", ":")).encode("utf-8"))
+                if byte_count + batch_bytes > BASELINE_MAX_BYTES:
+                    raise BaselineBudgetExceeded(
+                        f"BASELINE_BUDGET_EXCEEDED: 十五天历史基线超过 {BASELINE_MAX_BYTES // (1024 * 1024)} MiB 准备预算，"
+                        "尚未发布；请减少回放设备或核查历史采样密度后重试"
+                    )
+                async with self.repo.connection.transaction():
+                    await self.repo.save_batch(document["replay_id"], batch)
+                batch_count += 1
+                row_count += len(outputs)
+                byte_count += batch_bytes
+                for row in outputs:
+                    device_id = row["device_id"]
+                    counts[device_id] += 1
+                    if row["quality"] == "valid":
+                        valid_counts[device_id] += 1
+                        histories[device_id].append(row["cleaned_value"])
+                        histories[device_id] = histories[device_id][-window:]
+                last = rows[-1]
+                after = (last["time"], last["device_id"])
+        document["baseline"] = {
+            "start_time": start.isoformat(), "end_time": config.start_time.isoformat(),
+            "end_exclusive": True, "sample_counts": counts, "valid_sample_counts": valid_counts,
+            "row_count": row_count, "batch_count": batch_count, "byte_count": byte_count,
+        }
+        document["histories"] = histories
+        if calibration is not None:
+            from app.mcp.analysis_service import bounded_calculation
+            document["monitoring_calibration"] = await bounded_calculation(calibration.fit)
+        document["warmed"] = True
+        async with self.repo.connection.transaction():
+            await self.repo.publish_baseline(document["replay_id"])
+            await self.repo.save(document)
+
     async def _create_investigations(self, document, changes, batch, *, create_tasks=True):
         from app.domain.schemas.operational_task import OperationalTaskCreate, TaskSource, TaskTrigger
         from app.infra.db.postgresql.operational_tasks import OperationalTaskRepository
@@ -309,9 +441,20 @@ class ScadaReplayService:
         tasks = OperationalTaskService(OperationalTaskRepository(self.repo.connection, self.context.project_id),
                                        self.context, permissions_for_context(self.context))
         for event in changes:
+            # Period diagnosis owns measurement notifications during manual steps.
+            # Retain rule events as evidence; quality alerts remain independent.
+            if (not create_tasks and event.get("category") != "data_quality"
+                    and (event["change"] == "raised" or event.get("notification_suppressed"))):
+                if event["change"] == "raised":
+                    event["batch_sequence"] = batch["sequence"]
+                    event["notification_suppressed"] = True
+                    document["rule_states"][event.get("state_key", event["rule_id"])]["event"].update(
+                        batch_sequence=batch["sequence"], notification_suppressed=True,
+                    )
+                continue
             if event["change"] == "raised":
                 event["batch_sequence"] = batch["sequence"]
-                document["rule_states"][event["rule_id"]]["event"]["batch_sequence"] = batch["sequence"]
+                document["rule_states"][event.get("state_key", event["rule_id"])]["event"]["batch_sequence"] = batch["sequence"]
             await alerts.push(TaskAlertEvent(
                 event_id=uuid5(NAMESPACE_URL, event["event_id"] + ":" + event["change"]),
                 alert_id=event["event_id"], producer_id=producer, event_type=event["change"],
@@ -345,10 +488,30 @@ class ScadaReplayService:
             )
             task = await tasks.create(command)
             event["task_id"] = str(task.task_id)
-            document["rule_states"][event["rule_id"]]["event"]["task_id"] = event["task_id"]
+            document["rule_states"][event.get("state_key", event["rule_id"])]["event"]["task_id"] = event["task_id"]
 
     async def _create_monitoring_task(self, document, monitoring, batch):
-        if monitoring["status"] != "suspected_burst":
+        from app.services.scada_monitoring_incidents import attach_incident, update_incident
+        incident, action = update_incident(document, monitoring, batch["sequence"])
+        if incident is None:
+            return
+        if incident.get("task_id"):
+            if monitoring["status"] == "suspected_burst" and not incident.get("burst_diagnosed"):
+                from uuid import UUID
+                from app.infra.db.postgresql.operational_tasks import OperationalTaskRepository
+                from app.services.operational_tasks import OperationalTaskService
+                tasks = OperationalTaskService(
+                    OperationalTaskRepository(self.repo.connection, self.context.project_id),
+                    self.context, permissions_for_context(self.context),
+                )
+                await tasks.advance_replay_investigation(UUID(incident["task_id"]), {
+                    "replay_id": document["replay_id"], "knowledge_cutoff": monitoring["knowledge_cutoff"],
+                    "batch_sequence": batch["sequence"],
+                }, uuid5(NAMESPACE_URL, incident["incident_id"] + ":burst-evidence"))
+                incident["burst_diagnosed"] = True
+            if action == "resolved" and incident.get("alert"):
+                await self._publish_monitoring_alert(document, incident, "resolved")
+            attach_incident(monitoring, incident, action)
             return
         from app.domain.schemas.operational_task import OperationalTaskCreate, TaskSource, TaskTrigger
         from app.infra.db.postgresql.operational_tasks import OperationalTaskRepository
@@ -358,29 +521,44 @@ class ScadaReplayService:
             self.context,
             permissions_for_context(self.context),
         )
-        external = (
-            f"scada_replay:{document['replay_id']}:monitor:"
-            f"{monitoring['period_end']}:{monitoring['rule_version']}"
-        )
+        external = incident["external_id"]
         pressure = monitoring["pressure"]
         flow = monitoring["outlet_flow"]
         detector = monitoring["burst_detection"]
+        score = detector.get("target_score")
+        score_text = f"{score:.4f}" if score is not None else "数据不足"
         summary = (
-            f"步长监测疑似爆管：爆管侦测得分 {detector['target_score']:.4f}；"
-            f"压力下降测点 {len(pressure['hit_device_ids'])} 个；"
-            f"出厂边界流量增加测点 {len(flow['hit_device_ids'])} 个。"
+            f"SCADA 数据诊断发现指标异常，进入异常调查与爆管诊断。"
+            f"爆管侦测得分 {score_text}；"
+            f"压力下降测点 {len(pressure.get('hit_device_ids', []))} 个；"
+            f"出厂边界流量增加测点 {len(flow.get('hit_device_ids', []))} 个。"
+            "任一指标异常即启动调查，加权分仅用于评估爆管嫌疑，不能据此提前确诊。"
         )
+        if monitoring.get("anomaly_signals"):
+            signals = monitoring["anomaly_signals"]
+            devices = {device for item in signals for device in item.get("device_ids", [])}
+            summary += f" 单项异常 {len(signals)} 项，涉及 {len(devices)} 个测点；明细通过冻结 SCADA 查询核验。"
+        if monitoring.get("weighted"):
+            weighted = monitoring["weighted"]
+            weights = weighted["weights"]
+            summary += (
+                f" 综合评分 {weighted['score']:.2f}，爆管嫌疑门槛 {weighted['threshold']:.2f}"
+                f"（侦测 {weights['burst_detection']:.0%}、压力 {weights['pressure']:.0%}、"
+                f"流量 {weights['outlet_flow']:.0%}）。"
+            )
+            if detector.get("score_threshold") is not None:
+                summary += f" 侦测动态阈值 {detector['score_threshold']:.4f}。"
         command = OperationalTaskCreate(
             request_id=uuid5(NAMESPACE_URL, external),
             task_type="event_investigation",
-            workflow="burst_report",
-            title=f"SCADA 回放疑似爆管 · {monitoring['period_end']}",
+            workflow="scada_anomaly",
+            title=f"SCADA 异常调查 · {monitoring['period_end']}",
             summary=summary,
             source=TaskSource(
                 kind="external_system",
                 external_id=external,
                 external_revision=1,
-                event_kind="scada_burst_suspected",
+                event_kind="scada_anomaly",
                 replay_context={
                     "replay_id": document["replay_id"],
                     "knowledge_cutoff": monitoring["knowledge_cutoff"],
@@ -390,8 +568,33 @@ class ScadaReplayService:
             trigger=TaskTrigger(
                 kind="event",
                 occurred_at=monitoring["period_end"],
-                detail="主动步长监测的多项指标相互支持，进入爆管调查；尚未现场确诊",
+                detail="SCADA 任一指标异常启动调查，核验数据后进行爆管诊断；异常不等于已确诊爆管",
             ),
         )
         task = await tasks.create(command)
-        monitoring["task_id"] = str(task.task_id)
+        incident["task_id"] = str(task.task_id)
+        incident["burst_diagnosed"] = monitoring["status"] == "suspected_burst"
+        incident["alert"] = {
+            "alert_id": incident["incident_id"],
+            "producer_id": "scada-replay:" + document["replay_id"],
+            "occurred_at": incident["started_at"],
+            "condition_id": f"operational-task:{task.task_id}",
+            "source": {"kind": "scada_replay", "replay_id": document["replay_id"],
+                       "rule_id": "scada_diagnosis", "batch_sequence": batch["sequence"]},
+            "title": "SCADA 指标异常", "description": "已启动调查，详细诊断见调查报告。", "severity": "warning",
+        }
+        await self._publish_monitoring_alert(document, incident, "raised")
+        attach_incident(monitoring, incident, action)
+
+    async def _publish_monitoring_alert(self, document, incident, change):
+        from app.domain.schemas.task_alert import TaskAlertEvent
+        from app.infra.db.postgresql.task_alerts import TaskAlertRepository
+        from app.services.task_alerts import TaskAlertService
+        alerts = TaskAlertService(
+            TaskAlertRepository(self.repo.connection, self.context.project_id), self.context,
+            internal_producer_id="scada-replay:" + document["replay_id"],
+        )
+        await alerts.push(TaskAlertEvent(
+            **incident["alert"], event_type=change,
+            event_id=uuid5(NAMESPACE_URL, incident["incident_id"] + ":" + change),
+        ))

app/services/scada_replay_monitoring.py

@@ -9,6 +9,8 @@ from uuid import NAMESPACE_URL, uuid5
 import pandas as pd
 
 from app.domain.schemas.scada_replay import ScadaReplayCreate
+from app.domain.schemas.project_settings import ScadaMonitoringSettings
+from app.services.scada_normal_profile import current_calibration, normal_profile_bounds
 from app.services.burst_detection import (
     TARGET_MU,
     TARGET_N_ESTIMATORS,
@@ -42,6 +44,7 @@ def _relative_indicator(
     period_end: datetime,
     *,
     direction: str,
+    calibration: dict | None = None,
 ) -> dict[str, Any]:
     details: list[dict[str, Any]] = []
     gaps: list[dict[str, str]] = []
@@ -54,30 +57,51 @@ def _relative_indicator(
         if rule is None:
             gaps.append({"device_id": device_id, "reason": "relative_change_rule_missing"})
             continue
-        valid = _valid_rows(rows, device_id)
-        baseline = [row for row in valid if _time(row["time"]) < period_start][-rule.window_samples:]
-        current = [row for row in valid if period_start <= _time(row["time"]) <= period_end]
-        if len(baseline) < rule.minimum_samples:
-            gaps.append({"device_id": device_id, "reason": "baseline_not_ready"})
+        profile = calibration.get(device_id) if calibration is not None else None
+        if calibration is not None and profile is None:
+            gaps.append({"device_id": device_id, "reason": "adaptive_history_not_ready"})
             continue
+        threshold = float(profile["threshold_percent"]) if profile else rule.threshold
+        valid = _valid_rows(rows, device_id)
+        current = [row for row in valid if period_start < _time(row["time"]) <= period_end]
+        latest = current[-1] if current else None
         if not current:
             gaps.append({"device_id": device_id, "reason": "period_reading_missing"})
             continue
+        baseline = ([row for row in valid if _time(row["time"]) < _time(latest["time"])]
+                    [-rule.window_samples:]) if latest else []
+        if len(baseline) < rule.minimum_samples:
+            gaps.append({"device_id": device_id, "reason": "baseline_not_ready"})
+            continue
         reference = median(float(row["cleaned_value"]) for row in baseline)
         if reference == 0:
             gaps.append({"device_id": device_id, "reason": "zero_baseline"})
             continue
-        latest = current[-1]
         value = float(latest["cleaned_value"])
         change = (value - reference) / abs(reference) * 100
-        hit = change <= -rule.threshold if direction == "decrease" else change >= rule.threshold
+        hit = change <= -threshold if direction == "decrease" else change >= threshold
+        bounds = normal_profile_bounds(profile.get("normal_profile"), _time(latest["time"])) if profile else None
+        if profile and bounds is None:
+            gaps.append({"device_id": device_id, "reason": "normal_phase_history_not_ready"})
+            continue
+        if bounds is not None:
+            hit = hit and not bounds[0] <= value <= bounds[1]
         details.append({
             "device_id": device_id,
             "observed_at": _time(latest["time"]).isoformat(),
             "baseline": reference,
+            "baseline_method": "recent_samples_median",
+            "baseline_start": _time(baseline[0]["time"]).isoformat(),
+            "baseline_end": _time(baseline[-1]["time"]).isoformat(),
+            "baseline_sample_count": len(baseline),
             "value": value,
+            "value_change": value - reference,
+            "unit": rule.unit,
             "change_percent": change,
-            "threshold_percent": rule.threshold,
+            "threshold_percent": threshold,
+            **({"normal_phase_lower": bounds[0], "normal_phase_upper": bounds[1]} if bounds else {}),
+            **({"threshold_method": profile["method"],
+                "configured_threshold_percent": rule.threshold} if profile else {}),
             "hit": hit,
         })
     return {
@@ -94,6 +118,8 @@ def _burst_detection(
     pressure_device_ids: list[str],
     interval_seconds: int,
     period_end: datetime,
+    *,
+    score_threshold: float = TARGET_SCORE_THRESHOLD,
 ) -> dict[str, Any]:
     if len(pressure_device_ids) < 5:
         return {"status": "sensor_coverage_insufficient", "hit": False,
@@ -112,7 +138,10 @@ def _burst_detection(
         common_times = set(values) if common_times is None else common_times & set(values)
     if not common_times:
         return {"status": "baseline_not_ready", "hit": False, "required_samples": count}
-    target = max(common_times)
+    target = period_end
+    if target not in common_times:
+        return {"status": "current_reading_missing", "hit": False,
+                "target_time": target.isoformat(), "latest_common_time": max(common_times).isoformat()}
     expected = [target - timedelta(seconds=interval_seconds * offset)
                 for offset in range(count - 1, -1, -1)]
     incomplete = [device_id for device_id, values in series_by_device.items()
@@ -142,10 +171,11 @@ def _burst_detection(
     score = float(latest["Score"])
     return {
         "status": "ready",
-        "hit": score <= TARGET_SCORE_THRESHOLD,
+        "hit": score <= score_threshold,
         "target_time": target.isoformat(),
         "target_score": score,
-        "score_threshold": TARGET_SCORE_THRESHOLD,
+        "score_interpretation": "lower_is_more_anomalous_not_probability",
+        "score_threshold": score_threshold,
         "sensor_count": len(pressure_device_ids),
         "reference_day_count": 14,
         "latest_sensor_rankings": result["summary"].get("latest_sensor_rankings", []),
@@ -185,17 +215,64 @@ def evaluate_monitoring_period(
     flow_ids = monitoring.get("outlet_flow_device_ids", [])
     selected = set(config.device_ids)
     role_gaps = sorted((set(pressure_ids) | set(flow_ids)) - selected)
-    if not pressure_ids or not flow_ids or role_gaps:
+    weighted_policy = monitoring.get("decision_policy") == "weighted"
+    if not weighted_policy and (not pressure_ids or not flow_ids or role_gaps):
         return {**base, "status": "insufficient_data", "detail": "监测角色为空或不属于本轮回放设备",
                 "role_gaps": role_gaps,
                 "pressure": {"status": "unconfigured", "hit": False},
                 "outlet_flow": {"status": "unconfigured", "hit": False},
                 "burst_detection": {"status": "not_run", "hit": False}}
-    pressure = _relative_indicator(rows, config, pressure_ids, period_start, period_end, direction="decrease")
-    outlet_flow = _relative_indicator(rows, config, flow_ids, period_start, period_end, direction="increase")
-    detector = _burst_detection(
-        rows, pressure_ids, int(document["monitoring_interval_seconds"]), period_end,
+    calibration = current_calibration(document, config.start_time) if weighted_policy else None
+    if config.start_time > period_end:
+        calibration = None
+    base["role_gaps"] = role_gaps
+    pressure = _relative_indicator(
+        rows, config, pressure_ids, period_start, period_end, direction="decrease",
+        calibration=(calibration or {}).get("pressure", {}) if weighted_policy else None,
     )
+    outlet_flow = _relative_indicator(
+        rows, config, flow_ids, period_start, period_end, direction="increase",
+        calibration=(calibration or {}).get("outlet_flow", {}) if weighted_policy else None,
+    ) if flow_ids else {"status": "unconfigured", "hit": False, "hit_device_ids": [], "devices": [], "gaps": []}
+    detector = {"status": "baseline_not_ready", "hit": False} if weighted_policy and (
+        not calibration or calibration.get("status") != "ready"
+    ) else _burst_detection(
+        rows, pressure_ids, int(document["monitoring_interval_seconds"]), period_end,
+        score_threshold=calibration["burst_detection"]["threshold"] if calibration else TARGET_SCORE_THRESHOLD,
+    )
+    if weighted_policy:
+        settings = ScadaMonitoringSettings.model_validate(monitoring)
+        weights = settings.weights.model_dump()
+        indicators = {"burst_detection": detector, "pressure": pressure, "outlet_flow": outlet_flow}
+        contributions = {key: weights[key] if indicator["status"] == "ready" and indicator["hit"] else 0.0
+                         for key, indicator in indicators.items()}
+        score = round(sum(contributions.values()), 10)
+        complete = pressure["status"] == detector["status"] == "ready"
+        diagnostic_missing = sorted({device for rule in config.rules if rule.kind == "relative_change"
+                                     for device in rule.device_ids}
+                                    - set((calibration or {}).get("diagnostic", {})))
+        if not complete:
+            status, detail = "insufficient_data", (
+                "旧回放缺少新版历史诊断校准，请停止后新建演示" if calibration is None
+                else "压力数据或爆管侦测基线不足，尚不能作出组合判定"
+            )
+        elif score >= settings.alert_threshold:
+            status, detail = "suspected_burst", "爆管相关指标加权达到疑似爆管门槛"
+        elif any(indicator["hit"] for indicator in indicators.values()):
+            status, detail = "attention", "部分指标异常，进入调查；综合结果尚未达到爆管嫌疑门槛"
+        elif diagnostic_missing:
+            status, detail = "insufficient_data", "部分测点缺少完整历史同期基线，不能判定全部指标正常"
+        else:
+            status, detail = "normal", "本周期爆管相关指标均未命中"
+        detail += f"（综合 {score:.2f} / 门槛 {settings.alert_threshold:.2f}，流量权重 {weights['outlet_flow']:.0%}）"
+        return {**base, "status": status, "detail": detail, "pressure": pressure,
+                "outlet_flow": outlet_flow, "burst_detection": detector,
+                "weighted": {"score": score, "threshold": settings.alert_threshold,
+                             "weights": weights, "contributions": contributions,
+                             "unavailable_indicators": [key for key, value in indicators.items() if value["status"] != "ready"],
+                             "calibration_cutoff": calibration["knowledge_cutoff"] if calibration else None,
+                             "diagnostic_unavailable_device_ids": diagnostic_missing,
+                             "missing_weight_policy": "zero_contribution_without_renormalization"}}
     complete = pressure["status"] == outlet_flow["status"] == detector["status"] == "ready"
     hits = (bool(detector["hit"]), bool(pressure["hit"]), bool(outlet_flow["hit"]))
     if not complete:

app/services/scada_replay_processing.py

@@ -6,6 +6,9 @@ from statistics import median
 from uuid import NAMESPACE_URL, uuid5
 import re
 from app.domain.schemas.scada_replay import ScadaReplayCreate
+from app.services.scada_normal_profile import (
+    DIAGNOSTIC_POLICY, current_calibration, normal_profile_bounds, weighted_diagnosis,
+)
 
 
 def prepare_cleaning(rows: list[dict], config: ScadaReplayCreate) -> list[dict]:
@@ -43,6 +46,8 @@ def detect_frame(document: dict, rows: list[dict], config: ScadaReplayCreate, at
     by_id = {row["device_id"]: row for row in rows}
     changes = []
     devices = {device["device_id"]: device for device in result.get("devices", [])}
+    weighted = weighted_diagnosis(result)
+    calibration = current_calibration(result, config.start_time) if weighted else None
     for device_id in config.device_ids:
         row = by_id.get(device_id)
         invalid = row is not None and row["quality"] != "valid"
@@ -75,7 +80,8 @@ def detect_frame(document: dict, rows: list[dict], config: ScadaReplayCreate, at
                 changes.append({**state["event"], "status": "resolved", "resolved_at": at.isoformat(), "change": "resolved"})
                 state["event"] = None
     for rule in config.rules:
-        state = states.setdefault(rule.rule_id, {"hits": 0, "recoveries": 0, "event": None})
+        state_key = f"{rule.rule_id}:{DIAGNOSTIC_POLICY}" if weighted and rule.kind == "relative_change" else rule.rule_id
+        state = states.setdefault(state_key, {"hits": 0, "recoveries": 0, "event": None})
         hits, usable, reasons = [], [], []
         for device_id in rule.device_ids:
             row = by_id.get(device_id)
@@ -86,6 +92,17 @@ def detect_frame(document: dict, rows: list[dict], config: ScadaReplayCreate, at
                 triggered = ((rule.lower is not None and value < rule.lower)
                              or (rule.upper is not None and value > rule.upper))
                 reason = f"{device_id}={value} {rule.unit}，超出配置范围"
+            elif weighted:
+                # The percentage in the compatibility rule is not a diagnosis
+                # threshold. Learn each device's normal same-phase range first.
+                profile = (calibration or {}).get("diagnostic", {}).get(device_id)
+                bounds = normal_profile_bounds(profile, at)
+                if bounds is None:
+                    continue
+                low, high = bounds
+                triggered = (value < low or value > high if rule.direction == "both"
+                             else value > high if rule.direction == "increase" else value < low)
+                reason = f"{device_id}={value} {rule.unit}，超出冻结正常同时段范围 [{low:.4f}, {high:.4f}] {rule.unit}"
             else:
                 previous = histories.get(device_id, [])[-rule.window_samples:]
                 if len(previous) < rule.minimum_samples:
@@ -106,10 +123,12 @@ def detect_frame(document: dict, rows: list[dict], config: ScadaReplayCreate, at
             state["hits"] += 1
             state["recoveries"] = 0
             if state["event"] is None and state["hits"] >= rule.consecutive_samples:
-                event_id = str(uuid5(NAMESPACE_URL, f"{result['replay_id']}:{rule.rule_id}:{at.isoformat()}"))
+                event_id = str(uuid5(NAMESPACE_URL, f"{result['replay_id']}:{state_key}:{at.isoformat()}"))
                 event = {"event_id": event_id, "rule_id": rule.rule_id, "device_ids": hits,
                          "occurred_at": at.isoformat(), "last_observed_at": at.isoformat(),
                          "status": "active", "detail": "；".join(reasons)[:2000], "task_id": None}
+                if weighted and rule.kind == "relative_change":
+                    event.update(diagnostic_policy=DIAGNOSTIC_POLICY, state_key=state_key)
                 state["event"] = event
                 changes.append({**event, "change": "raised"})
                 result["anomaly_count"] = result.get("anomaly_count", 0) + 1
@@ -128,7 +147,9 @@ def detect_frame(document: dict, rows: list[dict], config: ScadaReplayCreate, at
         # Incomplete frames cannot count as a consecutive hit/recovery.
         elif not usable and any(device in by_id for device in rule.device_ids):
             state["hits"] = state["recoveries"] = 0
-    frozen = {device for rule in config.rules if states[rule.rule_id]["event"]
+    frozen = {device for rule in config.rules if states[
+                  f"{rule.rule_id}:{DIAGNOSTIC_POLICY}" if weighted and rule.kind == "relative_change" else rule.rule_id
+              ]["event"]
               for device in rule.device_ids}
     for row in rows:
         if row["quality"] == "valid" and row["device_id"] not in frozen:

app/services/scada_replay_queries.py

@@ -12,14 +12,23 @@ def check_replay_window(context, start, end):
         raise HTTPException(422, f"回放查询需要明确时间范围，结束不得晚于 {cap.isoformat()}")
 
 
+def check_snapshot_start(run, start):
+    replay_start = datetime.fromisoformat(run["start_time"])
+    if start < replay_start:
+        baseline = run.get("baseline")
+        if not baseline:
+            raise HTTPException(422, "本轮回放未冻结历史对照，请新建回放后查询历史基线")
+        if start < datetime.fromisoformat(baseline["start_time"]):
+            raise HTTPException(422, "查询起点早于本轮已冻结的历史基线")
+
+
 async def replay_snapshot_rows(connection, project_id, context, device_ids, start, end, limit, offset=0):
     check_replay_window(context, start, end)
     repository = ScadaReplayRepository(connection, project_id)
     run = await repository.get(context["replay_id"])
     if not set(device_ids) <= set(run["config"]["device_ids"]):
         raise HTTPException(403, "设备或回放不属于当前调查")
-    if start < datetime.fromisoformat(run["start_time"]):
-        raise HTTPException(422, "当前调查证据接口仅返回本轮已处理观测，回放前基线已用于后台初始化")
+    check_snapshot_start(run, start)
     rows = await repository.detected_outputs(
         context["replay_id"], device_ids, start, end, context["batch_sequence"],
         limit=limit, offset=offset,
@@ -40,9 +49,7 @@ async def replay_burst_location_inputs(connection, project_id, context, args):
     repository = ScadaReplayRepository(connection, project_id)
     run = await repository.get(context["replay_id"])
     if run:
-        replay_start = datetime.fromisoformat(run["start_time"])
-        if args.scada_burst_start < replay_start or args.scada_normal_start < replay_start:
-            raise HTTPException(422, "定位窗口必须位于本轮已处理回放范围内")
+        check_snapshot_start(run, min(args.scada_burst_start, args.scada_normal_start))
         devices = {item["device_id"]: item for item in run.get("devices", [])}
         monitoring = run.get("monitoring_config") or {}
         pressure_devices = monitoring.get("pressure_device_ids", [])