feat(task): route validation status to review states

app-instance/backend/beaver/services/agent_service.py

@@ -854,7 +854,19 @@ class AgentService:
                 provider_bundle=provider_bundle,
             )
             latest_validation = validation
-            task = task_service.record_validation(task.task_id, result.run_id, validation)
+            has_usable_answer = bool(result.output_text.strip()) and (
+                "Tool loop stopped after reaching the configured iteration limit." not in result.output_text
+            )
+            task = task_service.record_validation(
+                task.task_id,
+                result.run_id,
+                validation,
+                final_attempt=(
+                    attempt_index == 2
+                    or validation.status in {"accepted", "insufficient_evidence", "validator_error"}
+                ),
+                has_usable_answer=has_usable_answer,
+            )
             run_memory_store.update_run_record(result.run_id, validation_result=validation.to_dict())
             session_manager.update_latest_assistant_event_payload(
                 result.session_id,
@@ -865,6 +877,23 @@ class AgentService:
                     "validation_status": "passed" if validation.accepted else "failed",
                 },
             )
+            validation_debug = {
+                "evidence_run_ids": [
+                    item.run_id for item in [evidence_packet.main_run, *evidence_packet.team_runs] if item is not None
+                ],
+                "evidence_session_ids": [
+                    item.session_id
+                    for item in [evidence_packet.main_run, *evidence_packet.team_runs]
+                    if item is not None
+                ],
+                "tool_result_count": sum(
+                    len(item.tool_results)
+                    for item in [evidence_packet.main_run, *evidence_packet.team_runs]
+                    if item is not None
+                ),
+                "evidence_length": len(evidence_text),
+            }
+            retry_scheduled = validation.status == "rejected" and attempt_index == 1
             session_manager.append_message(
                 result.session_id,
                 run_id=result.run_id,
@@ -874,17 +903,18 @@ class AgentService:
                     "task_id": task.task_id,
                     "attempt_index": attempt_index,
                     "validation_result": validation.to_dict(),
-                    "retry_scheduled": not validation.accepted and attempt_index == 1,
+                    "validation_debug": validation_debug,
+                    "retry_scheduled": retry_scheduled,
                 },
                 content=validation.recommended_revision_prompt or None,
                 context_visible=False,
             )
-            if not validation.accepted and attempt_index == 1:
+            if retry_scheduled:
                 session_manager.set_run_context_visible(result.session_id, result.run_id, False)
             result.task_id = task.task_id
             result.task_status = task.status
             result.validation_result = validation.to_dict()
-            if validation.accepted or attempt_index == 2:
+            if not retry_scheduled:
                 return result
 
         if last_result is None:  # pragma: no cover - defensive

app-instance/backend/beaver/tasks/service.py

@@ -110,10 +110,30 @@ class TaskService:
         self._event(task, "run_completed", run_id=run_id, payload={"skill_names": skill_names or []})
         return task
 
-    def record_validation(self, task_id: str, run_id: str, validation: ValidationResult) -> TaskRecord:
+    def record_validation(
+        self,
+        task_id: str,
+        run_id: str,
+        validation: ValidationResult,
+        *,
+        final_attempt: bool = True,
+        has_usable_answer: bool = True,
+    ) -> TaskRecord:
         task = self._require(task_id)
-        task.status = "awaiting_feedback"
-        task.updated_at = self._now()
+        now = self._now()
+        if validation.status == "accepted":
+            task.status = "awaiting_feedback"
+        elif validation.status in {"insufficient_evidence", "validator_error"}:
+            task.status = "needs_review"
+        elif validation.status == "rejected" and not final_attempt:
+            task.status = "needs_revision"
+        elif validation.status == "rejected" and has_usable_answer:
+            task.status = "needs_review"
+        else:
+            task.status = "failed"
+            task.closed_at = now
+            task.close_reason = "automatic validation rejected the final attempt"
+        task.updated_at = now
         task.validation_result = validation.to_dict()
         self.store.upsert_task(task)
         self._event(task, "validated", run_id=run_id, payload=validation.to_dict())

app-instance/backend/beaver/tasks/validation.py

@@ -45,13 +45,13 @@ class ValidationService:
                 )
             except Exception as exc:
                 return ValidationResult(
-                    passed=False,
+                    status="validator_error",
                     score=0.0,
                     issues=[f"Validator failed: {exc}"],
-                    missing_requirements=["A valid automatic validation result is required before accepting the task."],
+                    evidence_gaps=["Automatic validation failed before producing a reliable decision."],
+                    missing_requirements=["User review is required because automatic validation failed."],
                     recommended_revision_prompt=(
-                        "Review the task result again because automatic validation failed, "
-                        "then provide a corrected final answer that explicitly satisfies the task goal."
+                        "Review the answer and evidence, then decide whether to revise or accept it."
                     ),
                     validator="llm_error",
                 )
@@ -96,11 +96,19 @@ class ValidationService:
             temperature=0.0,
         )
         payload = self._parse_json_object(response.content or "")
+        status = payload.get("status")
+        if status not in {"accepted", "rejected", "insufficient_evidence", "validator_error"}:
+            status = (
+                "accepted"
+                if payload.get("passed") and float(payload.get("score", 0.0) or 0.0) >= 0.75
+                else "rejected"
+            )
         return ValidationResult(
-            passed=bool(payload.get("passed")),
+            status=status,
             score=max(0.0, min(1.0, float(payload.get("score", 0.0) or 0.0))),
             issues=[str(item) for item in payload.get("issues") or []],
             missing_requirements=[str(item) for item in payload.get("missing_requirements") or []],
+            evidence_gaps=[str(item) for item in payload.get("evidence_gaps") or []],
             recommended_revision_prompt=str(payload.get("recommended_revision_prompt") or ""),
             validator="llm",
         )

app-instance/backend/tests/unit/test_task_mode_feedback.py

@@ -779,6 +779,45 @@ def test_task_mode_team_failure_still_uses_main_synthesis(tmp_path: Path) -> Non
     assert "user-visible fallback answer" in main_provider.calls[0]["messages"][0]["content"]
 
 
+def test_insufficient_evidence_moves_task_to_needs_review(tmp_path: Path) -> None:
+    service = AgentService(
+        loader=EngineLoader(
+            workspace=tmp_path,
+            task_execution_planner=_single_planner(),
+            validation_service=StubValidationService(
+                [
+                    ValidationResult(
+                        status="insufficient_evidence",
+                        score=0.4,
+                        evidence_gaps=["source missing"],
+                        validator="test",
+                    )
+                ]
+            ),
+        )
+    )
+
+    result = asyncio.run(
+        service.process_direct(
+            "answer with uncertain evidence",
+            session_id="web:needs-review",
+            provider_bundle=_bundle("possible answer"),
+        )
+    )
+    loaded = service.create_loop().boot()
+    task = loaded.task_service.get_task(result.task_id)
+    events = loaded.session_manager.get_run_event_records(result.session_id, result.run_id)
+    validation_event = next(event for event in events if event.event_type == "task_validation_snapshotted")
+
+    assert task is not None
+    assert task.status == "needs_review"
+    assert task.requires_user_action is True
+    assert task.is_execution_active is False
+    assert validation_event.event_payload["validation_result"]["status"] == "insufficient_evidence"
+    assert validation_event.event_payload["retry_scheduled"] is False
+    assert validation_event.event_payload["validation_debug"]["tool_result_count"] >= 0
+
+
 def test_task_mode_team_retry_hides_first_synthesis_run(tmp_path: Path) -> None:
     main_provider = StubProvider(
         [
@@ -890,5 +929,6 @@ def test_llm_validator_parse_failure_is_not_accepted(tmp_path: Path) -> None:
         )
 
     assert validation.accepted is False
+    assert validation.status == "validator_error"
     assert validation.validator == "llm_error"
     assert validation.issues