feat(backend): upgrade to the Gemini 3.x models

backend/.env.example

@@ -27,7 +27,7 @@ GOOGLE_APPLICATION_CREDENTIALS=keys/credentials.json
 # (e.g. us-central1) — these models are not published in the global publisher catalog.
 VERTEX_API_EMBEDDINGS_REGION=us-central1
 # Region for generative-AI calls (Gemini etc.). Can be a regional location or "global".
-VERTEX_API_GEN_AI_REGION=us-central1
+VERTEX_API_GEN_AI_REGION=global
 EMBEDDINGS_SERVICE_NAME=GOOGLE-VERTEX-AI
 # See https://cloud.google.com/vertex-ai/generative-ai/docs/embeddings/get-text-embeddings#supported-models e.g.:
 #   text-embedding-005
@@ -104,11 +104,13 @@ LLM_PROVIDER=gemini
 #   LLM_REASONING_MODEL       — classifiers, routing and extraction that need good reasoning
 #   LLM_DEEP_REASONING_MODEL  — the strongest reasoning, e.g. evaluation judges
 # When a tier is unset:
-#   gemini:    uses its own default (gemini-2.5-flash-lite / gemini-2.5-flash / gemini-2.5-pro)
+#   gemini:    uses its own default (gemini-3.5-flash-lite / gemini-3.8-flash / gemini-3.1-pro-preview)
 #   anthropic: falls back to the next tier down that is set
 #              (LLM_DEEP_REASONING_MODEL -> LLM_REASONING_MODEL -> LLM_DEFAULT_MODEL), then to claude-sonnet-4-6
 #   ollama:    same as anthropic, then to qwen2.5:7b
 # To use one model for everything with a non-Gemini provider, set only LLM_DEFAULT_MODEL.
+# Gemini models are called through the google-genai Interactions API, which only supports
+# gemini-3.5-flash-lite, gemini-3.8-flash and gemini-3.1-pro-preview (the 2.5 models are rejected).
 # LLM_DEFAULT_MODEL=
 # LLM_REASONING_MODEL=
 # LLM_DEEP_REASONING_MODEL=

backend/README.md

@@ -567,13 +567,14 @@ For more details, see [app/agent/preference_elicitation_agent/README.md](app/age
 ### AI & Embeddings
 
 - `VERTEX_API_EMBEDDINGS_REGION` - Vertex AI region for embedding models (must be a regional location, e.g. `us-central1`)
-- `VERTEX_API_GEN_AI_REGION` - Vertex AI region for generative-AI calls (regional or `global`, default: `us-central1`)
+- `VERTEX_API_GEN_AI_REGION` - Vertex AI region for generative-AI calls (regional or `global`, default: `global`). The default and reasoning Gemini 3.x models are only available in `global`
 - `EMBEDDINGS_SERVICE_NAME` - Embeddings service provider
 - `EMBEDDINGS_MODEL_NAME` - Model for generating embeddings
 - `LLM_PROVIDER` - LLM backend for all agents: `gemini` (default), `anthropic` or `ollama`
-- `LLM_DEFAULT_MODEL` - Model for the default tier (gemini default: `gemini-2.5-flash-lite`)
-- `LLM_REASONING_MODEL` - Model for the reasoning tier (gemini default: `gemini-2.5-flash`)
-- `LLM_DEEP_REASONING_MODEL` - Model for the deep reasoning tier (gemini default: `gemini-2.5-pro`)
+- `LLM_DEFAULT_MODEL` - Model for the default tier (gemini default: `gemini-3.5-flash-lite`)
+- `LLM_REASONING_MODEL` - Model for the reasoning tier (gemini default: `gemini-3.8-flash`)
+- `LLM_DEEP_REASONING_MODEL` - Model for the deep reasoning tier (gemini default: `gemini-3.1-pro-preview`)
+  - Gemini models are called through the google-genai Interactions API, which only supports `gemini-3.5-flash-lite`, `gemini-3.8-flash` and `gemini-3.1-pro-preview` (the 2.5 models are rejected)
   - For `anthropic` and `ollama`, an unset tier falls back to the next tier down that is set (`LLM_DEEP_REASONING_MODEL` → `LLM_REASONING_MODEL` → `LLM_DEFAULT_MODEL`), then to the provider default (`claude-sonnet-4-6` / `qwen2.5:7b`)
 - `ANTHROPIC_API_KEY` - Anthropic API key, required when `LLM_PROVIDER=anthropic`
 - `OLLAMA_BASE_URL` - Ollama server URL when `LLM_PROVIDER=ollama` (default: `http://localhost:11434`)

backend/app/agent/config.py

@@ -2,9 +2,9 @@
 from typing import Literal, Final
 
 Model = Literal[
-    "gemini-2.5-flash-lite",
-    "gemini-2.5-flash",
-    "gemini-2.5-pro",
+    "gemini-3.5-flash-lite",
+    "gemini-3.8-flash",
+    "gemini-3.1-pro-preview",
 ]
 
 
@@ -39,10 +39,12 @@ class ModelTier(StrEnum):
 
 
 GEMINI_DEFAULT_MODELS: Final[dict[ModelTier, Model]] = {
-    ModelTier.DEFAULT: "gemini-2.5-flash-lite",
-    ModelTier.REASONING: "gemini-2.5-flash",
-    ModelTier.DEEP_REASONING: "gemini-2.5-pro",
+    ModelTier.DEFAULT: "gemini-3.5-flash-lite",
+    ModelTier.REASONING: "gemini-3.8-flash",
+    ModelTier.DEEP_REASONING: "gemini-3.1-pro-preview",
 }
 """
 The Gemini model used for each tier when the tier's environment variable is not set.
+The Gemini 3.x models are only served from the Vertex AI `global` location, see `VERTEX_API_GEN_AI_REGION`.
+These are models the google-genai Interactions API supports, it rejects the Gemini 2.5 models.
 """

backend/app/app_config.py

@@ -143,22 +143,22 @@ class ApplicationConfig(BaseModel):
     llm_default_model: Optional[str] = None
     """
     The model for the default tier (ModelTier.DEFAULT) of the selected provider.
-    When unset, each provider uses its own default (gemini: gemini-2.5-flash-lite,
+    When unset, each provider uses its own default (gemini: gemini-3.5-flash-lite,
     anthropic: claude-sonnet-4-6, ollama: qwen2.5:7b).
     Set via LLM_DEFAULT_MODEL environment variable.
     """
 
     llm_reasoning_model: Optional[str] = None
     """
     The model for the reasoning tier (ModelTier.REASONING) of the selected provider.
-    When unset, gemini uses gemini-2.5-flash; other providers fall back to llm_default_model.
+    When unset, gemini uses gemini-3.8-flash; other providers fall back to llm_default_model.
     Set via LLM_REASONING_MODEL environment variable.
     """
 
     llm_deep_reasoning_model: Optional[str] = None
     """
     The model for the deep reasoning tier (ModelTier.DEEP_REASONING) of the selected provider.
-    When unset, gemini uses gemini-2.5-pro; other providers fall back to llm_reasoning_model, then llm_default_model.
+    When unset, gemini uses gemini-3.1-pro-preview; other providers fall back to llm_reasoning_model, then llm_default_model.
     Set via LLM_DEEP_REASONING_MODEL environment variable.
     """
 

backend/common_libs/llm/models_utils.py

@@ -24,8 +24,9 @@
 
 DEFAULT_VERTEX_API_GEN_AI_REGION = os.getenv("VERTEX_API_GEN_AI_REGION")
 if not DEFAULT_VERTEX_API_GEN_AI_REGION:
-    logging.warning("VERTEX_API_GEN_AI_REGION is not set. Using 'us-central1' as the default region.")
-    DEFAULT_VERTEX_API_GEN_AI_REGION = "us-central1"
+    # The Gemini 3.x models are only served from the global location, regional locations return a 404 for them.
+    logging.warning("VERTEX_API_GEN_AI_REGION is not set. Using 'global' as the default region.")
+    DEFAULT_VERTEX_API_GEN_AI_REGION = "global"
 else:
     logging.info("Default Vertex AI gen-AI region is %s", DEFAULT_VERTEX_API_GEN_AI_REGION)
 
@@ -167,7 +168,7 @@ def resolve_model_name(*, tier: ModelTier, provider: "LLMProvider") -> str:
             ModelTier.REASONING: app_config.llm_reasoning_model,
             ModelTier.DEEP_REASONING: app_config.llm_deep_reasoning_model,
         }
-        # Gemini falls back to its own 2.5 default for the tier rather than to another tier's model.
+        # Gemini falls back to its own default for the tier rather than to another tier's model.
         tiers_to_try = [tier] if provider == "gemini" else [tier, *_NON_GEMINI_FALLBACK_TIERS[tier]]
         for candidate_tier in tiers_to_try:
             if configured_models[candidate_tier]:

backend/common_libs/llm/test_model_resolution.py

@@ -27,19 +27,19 @@ def _set_llm_config(config: ApplicationConfig, *,
 
 class TestResolveModelName:
     @pytest.mark.parametrize("given_tier, expected_model", [
-        (ModelTier.DEFAULT, "gemini-2.5-flash-lite"),
-        (ModelTier.REASONING, "gemini-2.5-flash"),
-        (ModelTier.DEEP_REASONING, "gemini-2.5-pro"),
+        (ModelTier.DEFAULT, "gemini-3.5-flash-lite"),
+        (ModelTier.REASONING, "gemini-3.8-flash"),
+        (ModelTier.DEEP_REASONING, "gemini-3.1-pro-preview"),
     ])
-    def test_gemini_uses_gemini_2_5_defaults_when_no_model_is_set(self, setup_application_config,
+    def test_gemini_uses_gemini_defaults_when_no_model_is_set(self, setup_application_config,
                                                                   given_tier: ModelTier, expected_model: str):
         # GIVEN the gemini provider is configured without any tier models
         _set_llm_config(setup_application_config, provider="gemini")
 
         # WHEN the model name for the given tier is resolved
         actual_model = resolve_model_name(tier=given_tier, provider="gemini")
 
-        # THEN expect the gemini 2.5 default for the tier
+        # THEN expect the gemini default for the tier
         assert actual_model == expected_model
 
     @pytest.mark.parametrize("given_tier, expected_model", [
@@ -63,9 +63,9 @@ def test_uses_the_tier_model_when_it_is_set(self, setup_application_config,
         assert actual_model == expected_model
 
     @pytest.mark.parametrize("given_reasoning_model, given_tier, expected_model", [
-        (None, ModelTier.REASONING, "gemini-2.5-flash"),
-        (None, ModelTier.DEEP_REASONING, "gemini-2.5-pro"),
-        ("given-reasoning", ModelTier.DEEP_REASONING, "gemini-2.5-pro"),
+        (None, ModelTier.REASONING, "gemini-3.8-flash"),
+        (None, ModelTier.DEEP_REASONING, "gemini-3.1-pro-preview"),
+        ("given-reasoning", ModelTier.DEEP_REASONING, "gemini-3.1-pro-preview"),
     ])
     def test_gemini_does_not_fall_back_to_another_tier(self, setup_application_config,
                                                        given_reasoning_model: Optional[str],
@@ -78,7 +78,7 @@ def test_gemini_does_not_fall_back_to_another_tier(self, setup_application_confi
         # WHEN the model name for a tier without a model is resolved
         actual_model = resolve_model_name(tier=given_tier, provider="gemini")
 
-        # THEN expect the gemini 2.5 default for the tier, not another tier's model
+        # THEN expect the gemini default for the tier, not another tier's model
         assert actual_model == expected_model
 
     @pytest.mark.parametrize("given_tier", list(ModelTier))
@@ -144,9 +144,9 @@ def test_non_gemini_default_tier_does_not_fall_back_to_a_higher_tier(self, setup
         assert actual_model == expected_model
 
     @pytest.mark.parametrize("given_tier, expected_model", [
-        (ModelTier.DEFAULT, "gemini-2.5-flash-lite"),
-        (ModelTier.REASONING, "gemini-2.5-flash"),
-        (ModelTier.DEEP_REASONING, "gemini-2.5-pro"),
+        (ModelTier.DEFAULT, "gemini-3.5-flash-lite"),
+        (ModelTier.REASONING, "gemini-3.8-flash"),
+        (ModelTier.DEEP_REASONING, "gemini-3.1-pro-preview"),
     ])
     def test_ignores_the_models_of_another_provider(self, setup_application_config,
                                                     given_tier: ModelTier, expected_model: str):
@@ -159,13 +159,13 @@ def test_ignores_the_models_of_another_provider(self, setup_application_config,
         # WHEN the model name for the given tier is resolved for gemini (e.g. a gemini-only call site)
         actual_model = resolve_model_name(tier=given_tier, provider="gemini")
 
-        # THEN expect the gemini 2.5 default for the tier
+        # THEN expect the gemini default for the tier
         assert actual_model == expected_model
 
     @pytest.mark.parametrize("given_tier, expected_model", [
-        (ModelTier.DEFAULT, "gemini-2.5-flash-lite"),
-        (ModelTier.REASONING, "gemini-2.5-flash"),
-        (ModelTier.DEEP_REASONING, "gemini-2.5-pro"),
+        (ModelTier.DEFAULT, "gemini-3.5-flash-lite"),
+        (ModelTier.REASONING, "gemini-3.8-flash"),
+        (ModelTier.DEEP_REASONING, "gemini-3.1-pro-preview"),
     ])
     def test_uses_the_defaults_when_the_application_config_is_not_set(self, given_tier: ModelTier,
                                                                       expected_model: str):
@@ -175,7 +175,7 @@ def test_uses_the_defaults_when_the_application_config_is_not_set(self, given_ti
         # WHEN the model name for the given tier is resolved
         actual_model = resolve_model_name(tier=given_tier, provider="gemini")
 
-        # THEN expect the gemini 2.5 default for the tier
+        # THEN expect the gemini default for the tier
         assert actual_model == expected_model
 
 
@@ -214,7 +214,7 @@ def test_explicit_model_name_is_used_as_is(self, setup_application_config):
 
 class TestGeminiGenerativeLLM:
     @pytest.mark.parametrize("given_config, expected_model", [
-        (LLMConfig(), "gemini-2.5-flash-lite"),
+        (LLMConfig(), "gemini-3.5-flash-lite"),
         (LLMConfig(model_tier=ModelTier.DEEP_REASONING), "given-deep-reasoning"),
         (LLMConfig(model_tier=ModelTier.DEEP_REASONING, language_model_name="given-explicit-model"),
          "given-explicit-model"),

backend/evaluation_tests/agent_director/llm_agent_director_scripted_user_test.py

@@ -104,7 +104,7 @@ async def agent_director_exec(caplog: LogCaptureFixture, test_case: ScriptedUser
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.repeat(3)
 async def test_user_says_all_the_time_yes(caplog: LogCaptureFixture,
                                           setup_agent_director: Awaitable[tuple[ConversationMemoryManager, Callable[
@@ -142,7 +142,7 @@ async def test_user_says_all_the_time_yes(caplog: LogCaptureFixture,
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.repeat(3)
 async def test_user_talks_about_occupations(caplog: LogCaptureFixture,
                                             setup_agent_director: Awaitable[tuple[ConversationMemoryManager, Callable[
@@ -206,7 +206,7 @@ class AgentState:
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.repeat(3)
 async def test_argentina_counseling_flow(caplog: LogCaptureFixture,
                                          setup_agent_director: Awaitable[tuple[ConversationMemoryManager, Callable[

backend/evaluation_tests/career_explorer_agent/career_explorer_agent_scripted_user_test.py

@@ -183,7 +183,7 @@ def career_explorer_config_with_sectors(setup_multi_locale_app_config):
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.parametrize("test_case", TEST_CASES, ids=[tc.name for tc in TEST_CASES])
 async def test_career_explorer_sector_content(
     evals_setup, setup_multi_locale_app_config, career_explorer_config_with_sectors, test_case: SectorContentTestCase
@@ -239,7 +239,7 @@ async def test_career_explorer_sector_content(
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.parametrize(
     "test_case", TEST_CASES_GENERAL_KNOWLEDGE, ids=[tc.name for tc in TEST_CASES_GENERAL_KNOWLEDGE]
 )

backend/evaluation_tests/career_explorer_agent/non_priority_sector_explorer_eval_test.py

@@ -54,7 +54,7 @@ def career_explorer_config_with_sectors(setup_multi_locale_app_config):
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.parametrize("test_case", NON_PRIORITY_EXPLORER_TEST_CASES, ids=[tc.name for tc in NON_PRIORITY_EXPLORER_TEST_CASES])
 async def test_non_priority_sector_explorer_web_search(
     evals_setup,

backend/evaluation_tests/career_explorer_agent/priority_sector_explorer_eval_test.py

@@ -104,7 +104,7 @@ def career_explorer_config_with_sectors(setup_multi_locale_app_config):
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.parametrize("test_case", PRIORITY_EXPLORER_TEST_CASES, ids=[tc.name for tc in PRIORITY_EXPLORER_TEST_CASES])
 async def test_priority_sector_explorer_rag_content(
     evals_setup,

backend/evaluation_tests/career_explorer_agent/sector_relevance_classifier_eval_test.py

@@ -70,7 +70,7 @@ def career_explorer_config_with_sectors(setup_multi_locale_app_config):
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.parametrize("test_case", CLASSIFIER_TEST_CASES, ids=[tc.name for tc in CLASSIFIER_TEST_CASES])
 async def test_sector_relevance_classifier(
     evals_setup,
@@ -116,7 +116,7 @@ async def test_sector_relevance_classifier(
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 async def test_classifier_extracts_multiple_sectors(
     evals_setup,
     career_explorer_config_with_sectors,

backend/evaluation_tests/career_readiness_agent/career_readiness_agent_scripted_user_test.py

@@ -246,7 +246,7 @@ def _save_conversation_output(
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.parametrize("test_case", TEST_CASES, ids=[tc.name for tc in TEST_CASES])
 async def test_career_readiness_scripted(evals_setup, setup_multi_locale_app_config,
                                          test_case: CareerReadinessTestCase):
@@ -285,7 +285,7 @@ async def test_career_readiness_scripted(evals_setup, setup_multi_locale_app_con
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.parametrize("test_case", EDGE_CASE_TEST_CASES, ids=[tc.name for tc in EDGE_CASE_TEST_CASES])
 async def test_career_readiness_edge_cases(evals_setup, setup_multi_locale_app_config,
                                             test_case: CareerReadinessEdgeCaseTestCase):

backend/evaluation_tests/collect_experiences_agent/_data_extraction_llm_test.py

@@ -999,7 +999,7 @@ class _TestCaseDataExtraction(CompassTestCase):
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.repeat(3)
 @pytest.mark.parametrize('test_case', get_test_cases_to_run(test_cases_data_extraction),
                          ids=[case.name for case in get_test_cases_to_run(test_cases_data_extraction)])

backend/evaluation_tests/collect_experiences_agent/collect_experiences_agent_scripted_user_test.py

@@ -67,7 +67,7 @@ async def collect_experiences_exec(caplog: LogCaptureFixture, test_case: Scripte
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.repeat(3)
 async def test_kenyan_user_multiple_experiences_transition_issues(caplog: LogCaptureFixture,
                                                                   setup_collect_experiences_agent: Awaitable[tuple[
@@ -201,7 +201,7 @@ async def test_stuck_after_recap_confirmation(caplog: LogCaptureFixture,
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.repeat(3)
 async def test_quick_reply_options_appear_at_yes_no_questions(caplog: LogCaptureFixture,
                                                               setup_collect_experiences_agent: Awaitable[tuple[

backend/evaluation_tests/collect_experiences_agent/collect_experiences_agent_simulated_user_test.py

@@ -21,7 +21,7 @@
 
 
 @pytest.mark.asyncio
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.repeat(3)
 @pytest.mark.parametrize('test_case', get_test_cases_to_run(test_cases),
                          ids=[case.name for case in get_test_cases_to_run(test_cases)])

backend/evaluation_tests/collect_experiences_agent/data_extraction/entity_extraction_tool_test.py

@@ -385,7 +385,7 @@ class EntityExtractionToolTestCase(CompassTestCase):
 
 @pytest.mark.asyncio
 @pytest.mark.repeat(3)
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.parametrize('test_case', get_test_cases_to_run(test_cases),
                          ids=[case.name for case in get_test_cases_to_run(test_cases)])
 async def test_entity_extraction_tool(test_case: EntityExtractionToolTestCase, caplog):

backend/evaluation_tests/collect_experiences_agent/data_extraction/intent_analyzer_llm_test.py

@@ -264,7 +264,7 @@ class IntentAnalyzerToolTestCase(CompassTestCase):
 
 @pytest.mark.asyncio
 @pytest.mark.repeat(3)
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.parametrize('test_case', get_test_cases_to_run(test_cases),
                          ids=[case.name for case in get_test_cases_to_run(test_cases)])
 async def test_intent_analyzer_tool(test_case: IntentAnalyzerToolTestCase, caplog):

backend/evaluation_tests/collect_experiences_agent/data_extraction/temporal_classifier_test.py

@@ -298,7 +298,7 @@ class TemporalAndWorkTypeClassifierToolTestCase(CompassTestCase):
 
 @pytest.mark.asyncio
 @pytest.mark.repeat(3)
-@pytest.mark.evaluation_test("gemini-2.5-flash-lite/")
+@pytest.mark.evaluation_test("gemini-3.5-flash-lite/")
 @pytest.mark.parametrize('test_case', get_test_cases_to_run(test_cases),
                          ids=[case.name for case in get_test_cases_to_run(test_cases)])
 async def test_temporal_and_work_type_classification(test_case: TemporalAndWorkTypeClassifierToolTestCase, caplog,