Refactor: Move backend files to backend/ directory and split .gitignore

backend/tests/test_query_analyzer_refinement.py

@@ -0,0 +1,103 @@
+import pytest
+from unittest.mock import MagicMock, patch
+from langchain_core.messages import HumanMessage, AIMessage
+from ea_chatbot.graph.nodes.query_analyzer import query_analyzer_node, QueryAnalysis
+from ea_chatbot.graph.state import AgentState
+
+@pytest.fixture
+def base_state():
+    return {
+        "messages": [],
+        "question": "",
+        "analysis": None,
+        "next_action": "",
+        "summary": ""
+    }
+
+@patch("ea_chatbot.graph.nodes.query_analyzer.get_llm_model")
+def test_refinement_coreference_from_history(mock_get_llm, base_state):
+    """
+    Test that the analyzer can resolve Year/State from history.
+    User asks "What about in NJ?" after a Florida 2024 query.
+    Expected: next_action = 'plan', NOT 'clarify' due to missing year.
+    """
+    state = base_state.copy()
+    state["messages"] = [
+        HumanMessage(content="Show me 2024 results for Florida"),
+        AIMessage(content="Here are the 2024 results for Florida...")
+    ]
+    state["question"] = "What about in New Jersey?"
+    state["summary"] = "The user is looking for 2024 election results."
+
+    mock_llm = MagicMock()
+    mock_get_llm.return_value = mock_llm
+    mock_structured = MagicMock()
+    mock_llm.with_structured_output.return_value = mock_structured
+    
+    # We expect the LLM to eventually return 'plan' because it sees the context.
+    # For now, if it returns 'clarify', this test should fail once we update the prompt to BE less strict.
+    mock_structured.invoke.return_value = QueryAnalysis(
+        data_required=["2024 results", "New Jersey"],
+        unknowns=[],
+        ambiguities=[],
+        conditions=["state=NJ", "year=2024"],
+        next_action="plan"
+    )
+
+    result = query_analyzer_node(state)
+    assert result["next_action"] == "plan"
+    assert "NJ" in str(result["analysis"]["conditions"])
+
+@patch("ea_chatbot.graph.nodes.query_analyzer.get_llm_model")
+def test_refinement_tolerance_for_missing_format(mock_get_llm, base_state):
+    """
+    Test that the analyzer doesn't flag missing output format or database name.
+    User asks "Give me a graph of turnout".
+    Expected: next_action = 'plan', even if 'format' or 'db' is not in query.
+    """
+    state = base_state.copy()
+    state["question"] = "Give me a graph of voter turnout in 2024 for Florida"
+
+    mock_llm = MagicMock()
+    mock_get_llm.return_value = mock_llm
+    mock_structured = MagicMock()
+    mock_llm.with_structured_output.return_value = mock_structured
+    
+    mock_structured.invoke.return_value = QueryAnalysis(
+        data_required=["voter turnout", "Florida"],
+        unknowns=[],
+        ambiguities=[],
+        conditions=["year=2024"],
+        next_action="plan"
+    )
+
+    result = query_analyzer_node(state)
+    assert result["next_action"] == "plan"
+    # Ensure no ambiguities were added by the analyzer itself (hallucinated requirement)
+    assert len(result["analysis"]["ambiguities"]) == 0
+
+@patch("ea_chatbot.graph.nodes.query_analyzer.get_llm_model")
+def test_refinement_enforces_voter_identity_clarification(mock_get_llm, base_state):
+    """
+    Test that 'track the same voter' still triggers clarification.
+    """
+    state = base_state.copy()
+    state["question"] = "Track the same voter participation in 2020 and 2024."
+
+    mock_llm = MagicMock()
+    mock_get_llm.return_value = mock_llm
+    mock_structured = MagicMock()
+    mock_llm.with_structured_output.return_value = mock_structured
+    
+    # We WANT it to clarify here because voter identity is not defined.
+    mock_structured.invoke.return_value = QueryAnalysis(
+        data_required=["voter participation"],
+        unknowns=[],
+        ambiguities=["Please define what fields constitute 'the same voter' (e.g. ID, or Name and DOB)."],
+        conditions=[],
+        next_action="clarify"
+    )
+
+    result = query_analyzer_node(state)
+    assert result["next_action"] == "clarify"
+    assert "identity" in str(result["analysis"]["ambiguities"]).lower() or "same voter" in str(result["analysis"]["ambiguities"]).lower()