dc_007 · Deployment: model artifact missing
Status: ✓ Passing · Last run: 2026-06-09T06:21:32 · Pipeline: deployment
· History: 1/1 runs passed (100%)
What was tested
The model artifact the deployment expects is gone — GCS path deleted, re-train misnamed the output, whatever. Deployment must halt cleanly with a clear "model artifact not found" message; it must NOT silently produce zero-prediction rows OR crash with a stack trace that buries the cause. This case is one of the few where the deployment SHOULD halt. The deployment_completed_cleanly checker is expect-halt-aware via planted.expect_halt.
What we planted in the data
- 1,000 entities
- Signal strength: 0.50 (sigmoid slope multiplier)
n_new_rows=100
What the system did
- Model selected:
DeploymentScoringRunner
What we checked — all assertions passed
| Status | Assertion | Detail |
|---|---|---|
| ✓ | deployment_completed_cleanly |
expected halt fired: RuntimeError: deployment failed: model artifact for case dc_007 not found at expected location. Re-train the model… |
| ✓ | pipeline_halted_with |
halted with text containing 'not found': 'RuntimeError: deployment failed: model artifact for case dc_007 not found at expected locatio… |
| ✓ | pipeline_halted_with |
halted with text containing 'not found': 'RuntimeError: deployment failed: model artifact for case dc_007 not found at expected locatio… |
How this could have gone wrong
(no assertion descriptions provided)
Why this case matters
When the model is gone, the right response is "halt loudly so the on-call sees it" — NOT "produce numbers that look like predictions but aren't." This case ensures the loud-halt path stays loud.
Reproducing
# from auto_insight_api/
python -m validation.v2 run dc_007 --pipeline deployment -v
- Case config:
validation/v2/cases/dc_007_model_missing.yaml - Data shape:
deployment_model_missing - Analytics type:
deployment