diff --git a/apps/api/src/api/v1/webhooks.py b/apps/api/src/api/v1/webhooks.py
index 20a846e4..3f4d9a4e 100644
--- a/apps/api/src/api/v1/webhooks.py
+++ b/apps/api/src/api/v1/webhooks.py
@@ -263,6 +263,7 @@ async def _try_auto_repair_background(
             playbook=decision.playbook,
             is_cold_start=decision.is_cold_start,
             similarity_score=decision.similarity_score,
+            run_post_verification=False,
         )
 
         logger.info(
diff --git a/apps/api/src/services/auto_repair_service.py b/apps/api/src/services/auto_repair_service.py
index 11a08f24..a03e7b3d 100644
--- a/apps/api/src/services/auto_repair_service.py
+++ b/apps/api/src/services/auto_repair_service.py
@@ -108,6 +108,7 @@ class IAutoRepairService(Protocol):
         self,
         incident: Incident,
         playbook: Playbook,
+        run_post_verification: bool = True,
     ) -> AutoRepairResult:
         """
         執行自動修復
@@ -378,6 +379,7 @@ class AutoRepairService:
         playbook: Playbook,
         is_cold_start: bool = False,
         similarity_score: float | None = None,
+        run_post_verification: bool = True,
     ) -> AutoRepairResult:
         """
         執行自動修復
@@ -579,10 +581,17 @@ class AutoRepairService:
                             error=str(_inner_e),
                         )
 
-                _vl_task = _asyncio.create_task(_verify_and_learn())
-                if hasattr(self, "_pending_tasks"):
-                    self._pending_tasks.add(_vl_task)
-                    _vl_task.add_done_callback(self._pending_tasks.discard)
+                if run_post_verification:
+                    _vl_task = _asyncio.create_task(_verify_and_learn())
+                    if hasattr(self, "_pending_tasks"):
+                        self._pending_tasks.add(_vl_task)
+                        _vl_task.add_done_callback(self._pending_tasks.discard)
+                else:
+                    logger.info(
+                        "auto_repair_service_post_verify_delegated",
+                        incident_id=incident.incident_id,
+                        playbook_id=playbook.playbook_id,
+                    )
             except Exception as _vl_e:
                 logger.warning("auto_repair_verifier_setup_failed", error=str(_vl_e))
 
diff --git a/apps/api/src/services/post_execution_verifier.py b/apps/api/src/services/post_execution_verifier.py
index 0ca0f4a9..e52c66cb 100644
--- a/apps/api/src/services/post_execution_verifier.py
+++ b/apps/api/src/services/post_execution_verifier.py
@@ -126,11 +126,25 @@ class PostExecutionVerifier:
             logger.warning("verifier_timeout", incident_id=incident_id)
             if snapshot:
                 await _update_snapshot(snapshot, {}, "timeout")
+            else:
+                await _persist_fallback_snapshot(
+                    incident=incident,
+                    post_state={},
+                    result="timeout",
+                    action_taken=action_taken,
+                )
             return "timeout"
         except Exception:
             logger.exception("verifier_collect_error", incident_id=incident_id)
             if snapshot:
                 await _update_snapshot(snapshot, {}, "failed")
+            else:
+                await _persist_fallback_snapshot(
+                    incident=incident,
+                    post_state={},
+                    result="failed",
+                    action_taken=action_taken,
+                )
             return "failed"
 
         # 3. 對比前後狀態
@@ -140,6 +154,13 @@ class PostExecutionVerifier:
         # 4. 更新 EvidenceSnapshot
         if snapshot:
             await _update_snapshot(snapshot, post_state, result)
+        else:
+            await _persist_fallback_snapshot(
+                incident=incident,
+                post_state=post_state,
+                result=result,
+                action_taken=action_taken,
+            )
 
         logger.info(
             "verifier_done",
@@ -492,6 +513,60 @@ async def _update_snapshot(
         logger.exception("verifier_snapshot_update_failed", snapshot_id=snapshot.snapshot_id)
 
 
+async def _persist_fallback_snapshot(
+    *,
+    incident: "Incident",
+    post_state: dict[str, Any],
+    result: str,
+    action_taken: str,
+) -> None:
+    """
+    Persist verifier outcome even when the pre-decision snapshot is unavailable.
+
+    Live T14 evidence showed auto_repair rows with verifier decisions in logs but
+    NULL incident_evidence.verification_result because verify(snapshot=None) had
+    no durable target. This fallback makes the verification gate auditable without
+    pretending a pre-execution baseline existed.
+    """
+    incident_id = _get_incident_id(incident)
+    try:
+        snapshot = EvidenceSnapshot(incident_id=incident_id)
+        snapshot.post_execution_state = post_state
+        snapshot.verification_result = result
+        snapshot.matched_playbook_id = _extract_playbook_id(action_taken)
+        snapshot.sensors_attempted = 1
+        snapshot.sensors_succeeded = 1 if post_state else 0
+        snapshot.mcp_health = {"post_execution_verifier": bool(post_state)}
+        snapshot.evidence_summary = (
+            "[PostExecutionVerifier] fallback verification snapshot; "
+            f"action={action_taken[:160]}; result={result}; "
+            "pre_execution_state=missing"
+        )
+        await snapshot.save()
+        logger.info(
+            "verifier_fallback_snapshot_saved",
+            incident_id=incident_id,
+            snapshot_id=snapshot.snapshot_id,
+            result=result,
+            action=action_taken,
+        )
+    except Exception:
+        logger.warning(
+            "verifier_fallback_snapshot_save_failed",
+            incident_id=incident_id,
+            result=result,
+            exc_info=True,
+        )
+
+
+def _extract_playbook_id(action_taken: str) -> str | None:
+    for prefix in ("auto_repair_playbook:", "auto_repair:"):
+        if action_taken.startswith(prefix):
+            playbook_id = action_taken.removeprefix(prefix).split(":", 1)[0].strip()
+            return playbook_id or None
+    return None
+
+
 # ─────────────────────────────────────────────────────────────────────────────
 # Singleton
 # ─────────────────────────────────────────────────────────────────────────────
diff --git a/apps/api/tests/test_learning_chain_e2e.py b/apps/api/tests/test_learning_chain_e2e.py
index 977bde76..e81e4b2d 100644
--- a/apps/api/tests/test_learning_chain_e2e.py
+++ b/apps/api/tests/test_learning_chain_e2e.py
@@ -233,6 +233,45 @@ async def test_auto_repair_success_triggers_verify_and_learn(monkeypatch):
     assert call["matched_playbook_id"] == playbook.playbook_id
 
 
+@pytest.mark.asyncio
+async def test_auto_repair_can_delegate_post_verification(monkeypatch):
+    """
+    webhook 路徑會自行 await PostExecutionVerifier；service 層需可跳過內部
+    fire-and-forget 驗證，避免同一個修復產生兩組驗證與 Telegram 升級。
+    """
+    stub_verifier = StubVerifier(result="success")
+    stub_learning = StubLearningService()
+
+    import src.services.post_execution_verifier as _pev_mod
+    monkeypatch.setattr(_pev_mod, "_verifier", stub_verifier)
+
+    import src.services.learning_service as _ls_mod
+    monkeypatch.setattr(_ls_mod, "_learning_service", stub_learning)
+
+    playbook = _make_playbook()
+    pb_service = StubPlaybookService()
+    pb_service.add_playbook(playbook)
+
+    service = AutoRepairService(
+        playbook_service=pb_service,
+        cooldown_checker=_no_cooldown,
+    )
+
+    incident = _make_incident()
+    result = await service.execute_auto_repair(
+        incident,
+        playbook,
+        run_post_verification=False,
+    )
+
+    assert result.success is True
+
+    await asyncio.sleep(0.05)
+
+    assert stub_verifier.calls == []
+    assert stub_learning.verification_calls == []
+
+
 @pytest.mark.asyncio
 async def test_auto_repair_failure_does_not_call_verifier(monkeypatch):
     """
diff --git a/apps/api/tests/test_post_execution_verifier.py b/apps/api/tests/test_post_execution_verifier.py
index bada10d8..b0ff82ac 100644
--- a/apps/api/tests/test_post_execution_verifier.py
+++ b/apps/api/tests/test_post_execution_verifier.py
@@ -208,6 +208,37 @@ class TestVerify:
 
         assert result == "failed"
 
+    @pytest.mark.asyncio
+    async def test_snapshot_missing_persists_fallback_verification(self):
+        """snapshot=None 也必須把 verification_result 寫成可稽核 fallback snapshot。"""
+        verifier = PostExecutionVerifier()
+        incident = _stub_incident()
+        persist = AsyncMock()
+
+        with patch.object(
+            verifier,
+            "_collect_post_state",
+            new=AsyncMock(return_value={"status": "Running"}),
+        ):
+            with patch(
+                "src.services.post_execution_verifier._persist_fallback_snapshot",
+                new=persist,
+            ):
+                result = await verifier.verify(
+                    incident=incident,
+                    snapshot=None,
+                    action_taken="auto_repair_playbook:PB-TEST",
+                    warmup_sec=0.0,
+                )
+
+        assert result == "success"
+        persist.assert_awaited_once()
+        call_kwargs = persist.await_args.kwargs
+        assert call_kwargs["incident"] is incident
+        assert call_kwargs["post_state"] == {"status": "Running"}
+        assert call_kwargs["result"] == "success"
+        assert call_kwargs["action_taken"] == "auto_repair_playbook:PB-TEST"
+
     @pytest.mark.asyncio
     async def test_collect_timeout_returns_timeout(self):
         """MCP 蒐集超時 → "timeout"，不 raise"""