V3 post step 2 refinement: add temperature & top-p settings to config (and .env), add prompt fields back to job table so that the prompt settings get frozen at runtime for all sources being processed.

tests/integration/test_pipeline_flow.py

@@ -73,10 +73,13 @@ class TestPipelineSuccessFlow:
             image_path,
             *,
             prompt_name="transcribe_document.md",
+            prompt_text=None,
+            temperature=None,
+            top_p=None,
             settings=None,
             provider=None,
         ) -> TranscriptionResult:
-            _ = (image_path, prompt_name, settings, provider)
+            _ = (image_path, prompt_name, prompt_text, temperature, top_p, settings, provider)
             return TranscriptionResult(
                 text="Pipeline transcript",
                 provider="openrouter",
@@ -102,10 +105,10 @@ class TestPipelineSuccessFlow:
         assert job is not None
         assert job.status == JobStatus.TRANSCRIBED
         assert any(job_source.raw_transcription == "Pipeline transcript" for job_source in job.job_sources)
-        assert any(job_source.prompt_name == "transcribe_document.md" for job_source in job.job_sources)
-        assert any(job_source.user_prompt is not None for job_source in job.job_sources)
-        assert any(job_source.temperature == 0.2 for job_source in job.job_sources)
-        assert any(job_source.top_p == 0.85 for job_source in job.job_sources)
+        assert job.prompt_name == "transcribe_document.md"
+        assert job.user_prompt is not None
+        assert job.temperature == 0.2
+        assert job.top_p == 0.85
         assert any(job_source.ai_metadata == {"finish_reason": "stop", "usage": {"total_tokens": 42}} for job_source in job.job_sources)
         assert any(
             job_source.raw_api_response == {"id": "resp_123", "choices": [{"message": {"content": "Pipeline transcript"}}]}
@@ -142,11 +145,14 @@ class TestPipelineSuccessFlow:
             image_path,
             *,
             prompt_name="transcribe_document.md",
+            prompt_text=None,
+            temperature=None,
+            top_p=None,
             settings=None,
             provider=None,
         ) -> TranscriptionResult:
             page_name = Path(image_path).name
-            _ = (prompt_name, settings, provider)
+            _ = (prompt_name, prompt_text, temperature, top_p, settings, provider)
             return TranscriptionResult(
                 text=f"Transcript for {page_name}",
                 provider="openrouter",
@@ -171,6 +177,7 @@ class TestPipelineSuccessFlow:
         assert all(job_source.status == JobSourceStatus.TRANSCRIBED for job_source in job.job_sources)
         assert all(job_source.raw_transcription for job_source in job.job_sources)
         assert all(job_source.source is not None and job_source.source.raw_transcription for job_source in job.job_sources)
+        assert job.prompt_name == "transcribe_document.md"
 
     @pytest.mark.asyncio
     async def test_worker_marks_partial_success_when_some_sources_fail(
@@ -202,12 +209,15 @@ class TestPipelineSuccessFlow:
             image_path,
             *,
             prompt_name="transcribe_document.md",
+            prompt_text=None,
+            temperature=None,
+            top_p=None,
             settings=None,
             provider=None,
         ) -> TranscriptionResult:
             nonlocal call_count
             call_count += 1
-            _ = (prompt_name, settings, provider)
+            _ = (image_path, prompt_name, prompt_text, temperature, top_p, settings, provider)
             if call_count == 2:
                 raise RuntimeError("simulated page failure")
             return TranscriptionResult(
@@ -279,11 +289,14 @@ class TestPipelineSuccessFlow:
             image_path,
             *,
             prompt_name="transcribe_document.md",
+            prompt_text=None,
+            temperature=None,
+            top_p=None,
             settings=None,
             provider=None,
         ) -> TranscriptionResult:
             nonlocal call_count
-            _ = (image_path, prompt_name, settings, provider)
+            _ = (image_path, prompt_name, prompt_text, temperature, top_p, settings, provider)
             call_count += 1
             return TranscriptionResult(
                 text="new transcript",
@@ -332,10 +345,13 @@ class TestPipelineFailureFlow:
             image_path,
             *,
             prompt_name="transcribe_document.md",
+            prompt_text=None,
+            temperature=None,
+            top_p=None,
             settings=None,
             provider=None,
         ) -> TranscriptionResult:
-            _ = (image_path, prompt_name, settings, provider)
+            _ = (image_path, prompt_name, prompt_text, temperature, top_p, settings, provider)
             raise RuntimeError("pipeline provider failure")
 
         monkeypatch.setattr(

tests/services/test_store.py

@@ -56,6 +56,8 @@ async def test_create_job_for_document_sorts_uploads_and_creates_links(async_ses
     assert created_job is not None
     assert created_job.provider == "openrouter"
     assert created_job.model == "test-model"
+    assert created_job.prompt_name == "transcribe_document.md"
+    assert created_job.user_prompt is not None
 
     sources = (
         await async_session.exec(
@@ -78,7 +80,6 @@ async def test_create_job_for_document_sorts_uploads_and_creates_links(async_ses
     job_sources = (await async_session.exec(select(JobSource).where(JobSource.job_id == result.job_id))).all()
     assert len(job_sources) == 2
     assert set(result.source_ids) == {job_source.source_id for job_source in job_sources}
-    assert {job_source.prompt_name for job_source in job_sources} == {None}
 
 
 @pytest.mark.asyncio
@@ -110,6 +111,10 @@ async def test_create_upload_job_stores_source_under_document_id_directory(async
     assert source.file_hash == "2c8648d103e3dd7ad87660da0f126a1443b6d21ac1bd3ec000c5e24e2373a90c"
     assert source.file_size_bytes == len(b"image-bytes")
 
+    created_job = await async_session.get(Job, result.job_id)
+    assert created_job is not None
+    assert created_job.prompt_name == "transcribe_document.md"
+
 
 def test_store_person_portrait_stores_file_under_person_id_directory(tmp_path):
     settings = Settings(openrouter_api_key="test-key", upload_dir=tmp_path)

tests/services/test_v2_crud.py

@@ -255,12 +255,10 @@ async def test_update_job_source_transcription_persists_provider_json_payloads(d
         raw_api_response=raw_payload,
         provider="openrouter",
         model="test-model",
-        prompt_name="transcribe_document.md",
     )
 
     stored_rows = await transcriptions.list_job_sources(job_id=job.id)
     assert len(stored_rows) == 1
     assert stored_rows[0].raw_transcription == "provider transcript"
-    assert stored_rows[0].prompt_name == "transcribe_document.md"
     assert stored_rows[0].ai_metadata == metadata
     assert stored_rows[0].raw_api_response == raw_payload

tests/services/test_workflows_reliability.py

@@ -65,8 +65,17 @@ class TestWorkflowReliability:
 
             loaded = await services.jobs.read_job(job_id=job.id, session=session)
 
-        async def _never_returns(image_path, *, prompt_name="transcribe_document.md", settings=None, provider=None):
-            _ = (image_path, prompt_name, settings, provider)
+        async def _never_returns(
+            image_path,
+            *,
+            prompt_name="transcribe_document.md",
+            prompt_text=None,
+            temperature=None,
+            top_p=None,
+            settings=None,
+            provider=None,
+        ):
+            _ = (image_path, prompt_name, prompt_text, temperature, top_p, settings, provider)
             raise TimeoutError("simulated provider timeout")
 
         monkeypatch.setattr("transcription.services.workflows.transcribe_document_image", _never_returns)

tests/ui/conftest.py

@@ -95,6 +95,7 @@ async def seed_job(app_client: tuple[FastAPI, TestClient]) -> Callable[..., Awai
                 retry_count=0,
                 provider="openrouter",
                 model="google/gemini-2.5-flash",
+                prompt_name="transcribe_document.md",
             )
             session.add(job)
             await session.flush()
@@ -121,7 +122,6 @@ async def seed_job(app_client: tuple[FastAPI, TestClient]) -> Callable[..., Awai
                             if transcription_text is not None
                             else JobSourceStatus.FAILED
                         ),
-                        prompt_name="transcribe_document.md",
                         raw_transcription=transcription_text,
                         error_detail=error_detail,
                     )