diff --git a/src/lai_automation/models.py b/src/lai_automation/models.py
index 7162518..4dfc726 100644
--- a/src/lai_automation/models.py
+++ b/src/lai_automation/models.py
@@ -12,6 +12,8 @@ class FailureReport:
     source: str
     category: str
     user_feedback: str | None
+    feedback_type: str | None
+    expected_behavior: str | None
     prompt: str | None
     response: str | None
     sql_query: str | None
@@ -29,6 +31,8 @@ class FailureReport:
             source=row["source"],
             category=row["category"],
             user_feedback=row.get("user_feedback"),
+            feedback_type=row.get("feedback_type"),
+            expected_behavior=row.get("expected_behavior"),
             prompt=row.get("prompt"),
             response=row.get("response"),
             sql_query=row.get("sql_query"),
diff --git a/src/lai_automation/prompt.py b/src/lai_automation/prompt.py
index e318cdc..10a6b3f 100644
--- a/src/lai_automation/prompt.py
+++ b/src/lai_automation/prompt.py
@@ -29,6 +29,8 @@ def redact(value: str | None) -> str:
 def build_task(report: FailureReport) -> str:
     prompt = redact(report.prompt)
     feedback = redact(report.user_feedback)
+    expected_behavior = redact(report.expected_behavior)
+    response = redact(report.response)
     error = redact(report.error_message)
     sql = redact(report.sql_query)
 
@@ -44,6 +46,7 @@ Rules:
 - Do not commit or push; the orchestrator owns Git publication.
 
 Failure category: {report.category}
+Feedback type: {report.feedback_type or 'not specified'}
 Source: {report.source}
 Model: {redact(str(report.context.get('model', 'unknown')))}
 Database type: {redact(str(report.context.get('database_type', 'unknown')))}
@@ -54,6 +57,12 @@ User request:
 User feedback:
 {feedback or '[not available]'}
 
+Expected behavior:
+{expected_behavior or '[not available]'}
+
+Observed answer:
+{response or '[not available]'}
+
 Observed error:
 {error or '[not available]'}
 
diff --git a/tests/test_prompt.py b/tests/test_prompt.py
index a700cdd..36a15d7 100644
--- a/tests/test_prompt.py
+++ b/tests/test_prompt.py
@@ -19,8 +19,10 @@ def test_task_contains_constraints_and_redacted_report() -> None:
         source="user_feedback",
         category="incorrect_answer",
         user_feedback="password=secret-value",
+        feedback_type="wrong_join",
+        expected_behavior="Join FK_Artikel with FK_AuftragPos and filter Auftragkennung = 3.",
         prompt="List locations for admin@example.com",
-        response=None,
+        response="The wrong product count was returned.",
         sql_query="SELECT * FROM company",
         error_message=None,
         context={"model": "mistral", "database_type": "customer"},
@@ -31,6 +33,9 @@ def test_task_contains_constraints_and_redacted_report() -> None:
     assert "admin@example.com" not in task
     assert "Do not commit or push" in task
     assert "never mock or stub the behavior being tested" in task
+    assert "Feedback type: wrong_join" in task
+    assert "Join FK_Artikel with FK_AuftragPos and filter Auftragkennung = 3." in task
+    assert "The wrong product count was returned." in task
 
 
 def test_redact_removes_extended_credentials() -> None:
diff --git a/tests/test_worker.py b/tests/test_worker.py
index c50bf35..73154b1 100644
--- a/tests/test_worker.py
+++ b/tests/test_worker.py
@@ -13,6 +13,8 @@ def test_dry_run_generates_task_without_agent_or_git() -> None:
         source="automatic",
         category="sql_execution",
         user_feedback=None,
+        feedback_type=None,
+        expected_behavior=None,
         prompt="List locations",
         response=None,
         sql_query=None,