V4.7 Phase 2: Evidence Model Simplification (part 2)

tests/integration/test_pipeline_flow.py

@@ -6,9 +6,12 @@ from uuid import uuid4
 
 import pytest
 from PIL import Image
+from sqlmodel import col
+from sqlmodel import select
 
 from transcription.config import Settings
 from transcription.db.models import Document
+from transcription.db.models import ExecutionAttempt
 from transcription.db.models import JobSourceStatus
 from transcription.db.models import JobStatus
 from transcription.providers.base import ProviderUsage
@@ -20,6 +23,15 @@ from transcription.services.store import create_job_for_document
 from transcription.services.workflows import advance_job
 
 
+async def _attempts_for_job(session, job) -> list[ExecutionAttempt]:
+    """Load execution attempts for a job; V4.7 moved evidence off JobSource."""
+    job_source_ids = [job_source.id for job_source in job.job_sources]
+    result = await session.exec(
+        select(ExecutionAttempt).where(col(ExecutionAttempt.job_source_id).in_(job_source_ids))
+    )
+    return list(result.all())
+
+
 def _jpeg_bytes(color: str = "white") -> bytes:
     output = io.BytesIO()
     Image.new("RGB", (2, 2), color=color).save(output, format="JPEG")
@@ -116,21 +128,24 @@ class TestPipelineSuccessFlow:
         assert processed is True
         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)
+        attempts = await _attempts_for_job(async_session, job)
+        assert any(attempt.raw_transcription == "Pipeline transcript" for attempt in attempts)
         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
+            attempt.normalized_metadata is not None
+            and attempt.normalized_metadata["finish_reason"] == "stop"
+            and attempt.normalized_metadata["usage"] == {"total_tokens": 42}
+            for attempt in attempts
         )
         assert any(
-            job_source.raw_api_response
+            attempt.sdk_response_snapshot
             == {"id": "resp_123", "choices": [{"message": {"content": "Pipeline transcript"}}]}
-            for job_source in job.job_sources
+            for attempt in attempts
         )
-        assert all(job_source.error_detail is None for job_source in job.job_sources)
+        assert all(attempt.error_detail is None for attempt in attempts)
 
     @pytest.mark.asyncio
     async def test_worker_transcribes_all_sources_for_multi_page_job(
@@ -191,7 +206,8 @@ class TestPipelineSuccessFlow:
         assert job.status == JobStatus.TRANSCRIBED
         assert len(job.job_sources) == 3
         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)
+        attempts = await _attempts_for_job(async_session, job)
+        assert all(attempt.raw_transcription for attempt in attempts)
         assert all(
             job_source.source is not None and job_source.source.raw_transcription for job_source in job.job_sources
         )
@@ -261,7 +277,8 @@ class TestPipelineSuccessFlow:
         assert len(job.job_sources) == 2
         statuses = {job_source.status for job_source in job.job_sources}
         assert statuses == {JobSourceStatus.TRANSCRIBED, JobSourceStatus.FAILED}
-        assert any(job_source.error_detail is not None for job_source in job.job_sources)
+        attempts = await _attempts_for_job(async_session, job)
+        assert any(attempt.error_detail is not None for attempt in attempts)
 
     @pytest.mark.asyncio
     async def test_worker_skips_already_transcribed_sources_on_resubmit(
@@ -293,9 +310,7 @@ class TestPipelineSuccessFlow:
         page_two = next(js for js in job.job_sources if js.source is not None and js.source.page_number == 2)
 
         page_one.status = JobSourceStatus.TRANSCRIBED
-        page_one.raw_transcription = "existing transcript"
         page_two.status = JobSourceStatus.PENDING
-        page_two.raw_transcription = None
         await services.sources.update_job_source(job_source=page_one, session=async_session)
         await services.sources.update_job_source(job_source=page_two, session=async_session)
         await services.jobs.update_job_state(job_id=job.id, status=JobStatus.QUEUED, session=async_session)
@@ -387,11 +402,10 @@ class TestPipelineFailureFlow:
         assert processed is True
         assert job is not None
         assert job.status == JobStatus.FAILED
-        assert all(job_source.raw_transcription is None for job_source in job.job_sources)
-        assert any(job_source.error_detail is not None for job_source in job.job_sources)
-        error_detail = next(
-            job_source.error_detail for job_source in job.job_sources if job_source.error_detail is not None
-        )
+        attempts = await _attempts_for_job(async_session, job)
+        assert all(attempt.raw_transcription is None for attempt in attempts)
+        assert any(attempt.error_detail is not None for attempt in attempts)
+        error_detail = next(attempt.error_detail for attempt in attempts if attempt.error_detail is not None)
         assert "pipeline provider failure" in error_detail
         assert "[internal_unexpected_error]" in error_detail
         assert "error_id=" in error_detail