refactor(tests): remove non-deterministic journey eval cases 4.2–4.5

Keep only 4.1 (first reply contains question) as automated eval.
Multi-turn cases (4.2–4.5) are non-deterministic and tested manually
with results tracked in Langfuse.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

tests/test_journey_v2.py

@@ -12,16 +12,17 @@ Unit tests (no LLM)
   4.6e  Session not found → done=True, agent_config=None
   4.6f  Nudge uses AGENT_CONFIG_START/END markers (not old PROMPT_TEMPLATE)
 
-Eval tests (real LLM + Langfuse scoring)
------------------------------------------
-Cases are defined in tests/fixtures/journey_v2/cases.yaml.
-Email HTML files live in tests/fixtures/journey_v2/data/.
-Use --journey-dir to point at a custom folder (same structure required).
+Eval test (real LLM + Langfuse scoring)
+----------------------------------------
+  4.1   Journey start explores directory → first reply contains a question
+
+Cases 4.2–4.5 (multi-turn conversations producing a full AgentConfig) are
+non-deterministic and tested manually — results tracked in Langfuse.
 
 Run:
     pytest tests/test_journey_v2.py -v
     pytest tests/test_journey_v2.py -v -k "4_6"          # unit only
-    pytest tests/test_journey_v2.py -v -k "eval"          # LLM evals only
+    pytest tests/test_journey_v2.py -v -k "eval"          # single LLM eval
     pytest tests/test_journey_v2.py -v --journey-dir /p   # custom fixtures
 """
 
@@ -170,57 +171,6 @@ def _evaluate_case(case: dict, reply: dict) -> tuple[float, str]:
         has_q = "?" in reply.get("message", "")
         return (1.0 if has_q else 0.0), f"first_reply_has_question={has_q}"
 
-    if case.get("expect_done") and not reply.get("done"):
-        return 0.0, "expected done=True but journey did not complete"
-
-    agent_config_raw = reply.get("agent_config")
-
-    if case.get("expect_valid_config"):
-        if not agent_config_raw:
-            return 0.0, "agent_config is None"
-        try:
-            parsed = AgentConfig.model_validate_json(agent_config_raw)
-            valid = len(parsed.content_types) > 0
-            return (1.0 if valid else 0.0), f"content_types={len(parsed.content_types)}"
-        except Exception as exc:
-            return 0.0, f"parse error: {exc}"
-
-    if case.get("expect_content_type_id"):
-        expected_id = case["expect_content_type_id"]
-        if not agent_config_raw:
-            return 0.0, "agent_config is None"
-        try:
-            parsed = AgentConfig.model_validate_json(agent_config_raw)
-            ids = [ct.id for ct in parsed.content_types]
-            found = expected_id in ids
-            return (1.0 if found else 0.0), f"content_type_ids={ids}, expected={expected_id}"
-        except Exception as exc:
-            return 0.0, f"parse error: {exc}"
-
-    if case.get("expect_extraction_contains"):
-        keyword = case["expect_extraction_contains"].lower()
-        if not agent_config_raw:
-            return 0.0, "agent_config is None"
-        try:
-            parsed = AgentConfig.model_validate_json(agent_config_raw)
-            if not parsed.content_types:
-                return 0.0, "no content_types in config"
-            prompt = parsed.content_types[0].extraction_prompt.lower()
-            found = keyword in prompt
-            return (1.0 if found else 0.0), f"keyword='{keyword}' in extraction_prompt={found}"
-        except Exception as exc:
-            return 0.0, f"parse error: {exc}"
-
-    if case.get("expect_global_rules"):
-        if not agent_config_raw:
-            return 0.0, "agent_config is None"
-        try:
-            parsed = AgentConfig.model_validate_json(agent_config_raw)
-            has_rules = len(parsed.global_rules) > 0
-            return (1.0 if has_rules else 0.0), f"global_rules={parsed.global_rules}"
-        except Exception as exc:
-            return 0.0, f"parse error: {exc}"
-
     return 1.0, "no specific assertion"