fix(navigation): stabilize GramPilot autonomous grid navigation and Qdrant persistence

- Hardened `NavigationKnowledge` and `TelepathicEngine` to prevent premature blacklisting on high-latency grid taps by returning `None` (inconclusive) when grid markers are still present.
- Updated `_execute_action` in `goap.py` to respect inconclusive verification states and avoid polluting the blacklist.
- Refined `Adaptive Snap` in `timing.py` to detect `ProgressBar` loading spinners, extending timeouts for slow connections.
- Prevented brittle Adaptive Snap `wobble` routines from firing while the bot is still trapped on the Explore Grid.
- Stabilized Qdrant collections by converting destructive delete-only wipes into safe `wipe_collection` routines that instantly recreate the index.
- Maintained 100% TDD pass-rate by aligning E2E grid navigation tests with the new inconclusive states.

tests/integration/test_llm_edge_inference.py

@@ -0,0 +1,42 @@
+import pytest
+from unittest.mock import patch, MagicMock
+from GramAddict.core.llm_provider import query_telepathic_llm
+
+def test_query_telepathic_llm_already_local():
+    # If the provided URL is local, it should NOT switch to fallback model
+    with patch("GramAddict.core.llm_provider.query_llm", return_value={"response": "test"}) as mock_query:
+        query_telepathic_llm(
+            model="llama3.2-vision",
+            url="http://localhost:11434/api/generate",
+            system_prompt="sys",
+            user_prompt="user",
+            use_local_edge=True
+        )
+        mock_query.assert_called_once()
+        args, kwargs = mock_query.call_args
+        assert kwargs["model"] == "llama3.2-vision"
+        assert kwargs["url"] == "http://localhost:11434/api/generate"
+
+def test_query_telepathic_llm_remote_with_local_edge():
+    # If the provided URL is remote, it SHOULD switch to fallback model when edge=True
+    
+    class MockArgs:
+        ai_fallback_model = "llama3.2:1b"
+        ai_fallback_url = "http://localhost:11434/api/generate"
+
+    class MockConfig:
+        args = MockArgs()
+
+    with patch("GramAddict.core.llm_provider.query_llm", return_value={"response": "test"}) as mock_query:
+        with patch("GramAddict.core.config.Config", return_value=MockConfig()):
+            query_telepathic_llm(
+                model="gpt-4o",
+                url="https://api.openai.com/v1/chat/completions",
+                system_prompt="sys",
+                user_prompt="user",
+                use_local_edge=True
+            )
+        mock_query.assert_called_once()
+        args, kwargs = mock_query.call_args
+        assert kwargs["model"] == "llama3.2:1b"
+        assert kwargs["url"] == "http://localhost:11434/api/generate"

tests/integration/test_qdrant_wipe.py

@@ -0,0 +1,49 @@
+import pytest
+from unittest.mock import patch, MagicMock
+
+from GramAddict.core.qdrant_memory import QdrantBase
+from GramAddict.core.telepathic_engine import TelepathicEngine
+
+@pytest.fixture(autouse=True)
+def mock_qdrant():
+    with patch('GramAddict.core.qdrant_memory.QdrantClient') as mq:
+        yield mq
+
+def test_qdrant_wipe_recreates_collection(mock_qdrant):
+    """
+    Tests that calling wipe_collection() on QdrantBase successfully calls 
+    delete_collection AND create_collection to prevent 404 errors.
+    """
+    mock_client = MagicMock()
+    mock_qdrant.return_value = mock_client
+    
+    # Missing collection creation during init
+    mock_client.collection_exists.return_value = False
+    base = QdrantBase("test_collection", vector_size=4)
+    
+    assert mock_client.create_collection.call_count == 1
+    
+    # Now call wipe_collection
+    base.wipe_collection()
+    
+    mock_client.delete_collection.assert_called_with("test_collection")
+    # create_collection should now have been called a 2nd time
+    assert mock_client.create_collection.call_count == 2
+
+def test_telepathic_engine_wipe_uses_wipe_collection(mock_qdrant):
+    """
+    Tests that TelepathicEngine.wipe() uses the safe wipe_collection method.
+    """
+    mock_client = MagicMock()
+    mock_qdrant.return_value = mock_client
+    
+    engine = TelepathicEngine()
+    
+    # Spy on the wipe_collection method
+    with patch.object(engine.embedding_helper, 'wipe_collection') as mock_emb_wipe, \
+         patch.object(engine.ui_memory, 'wipe_collection') as mock_ui_wipe:
+        
+        engine.wipe()
+        
+        mock_emb_wipe.assert_called_once()
+        mock_ui_wipe.assert_called_once()

tests/integration/test_situational_awareness.py

@@ -0,0 +1,28 @@
+import pytest
+from unittest.mock import MagicMock, patch
+from GramAddict.core.situational_awareness import SituationalAwarenessEngine, SituationType
+
+@pytest.fixture
+def sae():
+    device = MagicMock()
+    device.app_id = "com.instagram.android"
+    device.deviceV2.info = {"screenOn": True}
+    return SituationalAwarenessEngine(device)
+
+def test_perceive_normal_with_unknown_keyboard(sae):
+    # XML contains Instagram and some unknown keyboard
+    xml = """
+    <hierarchy>
+        <node package="com.instagram.android" resource-id="com.instagram.android:id/button" />
+        <node package="com.unknown.keyboard" resource-id="com.unknown.keyboard:id/key" />
+    </hierarchy>
+    """
+    
+    # We shouldn't call LLM for foreign app
+    with patch('GramAddict.core.llm_provider.query_telepathic_llm') as mock_llm:
+        # Let's mock ScreenMemoryDB to return NORMAL
+        with patch('GramAddict.core.qdrant_memory.ScreenMemoryDB.get_screen_type', return_value="NORMAL"):
+            res = sae.perceive(xml)
+            assert res == SituationType.NORMAL
+            # The LLM for foreign app should NOT have been called.
+            mock_llm.assert_not_called()

tests/unit/test_explore_grid_navigation.py

@@ -166,10 +166,10 @@ class TestVerifySuccessExploreGrid:
     just didn't register; it doesn't mean the mapping is wrong.
     """
 
-    def test_verify_success_returns_false_when_still_on_grid(self):
+    def test_verify_success_returns_none_when_still_on_grid(self):
         """
         If we tapped a grid item but the screen still shows the explore grid
-        (no feed markers), verify_success must return False.
+        (no feed markers), verify_success must return None (Inconclusive).
         """
         engine = TelepathicEngine()
         
@@ -192,7 +192,7 @@ class TestVerifySuccessExploreGrid:
         """
         
         result = engine.verify_success("first image in explore grid", still_on_grid_xml)
-        assert result is False, "verify_success should fail when still on explore grid"
+        assert result is None, "verify_success should be inconclusive (None) when still on explore grid"
 
     def test_verify_success_returns_true_when_post_opened(self):
         """