dc_008 · Deployment: high-volume batch (10× normal)
Status: ✓ Passing · Last run: 2026-06-09T06:21:50 · Pipeline: deployment
· History: 1/1 runs passed (100%)
What was tested
A backfill or replay pushed 5000 rows in a single fire instead of the usual ~100. Deployment must produce exactly 5000 predictions — no OOM, no truncation, no batching surprises.
What we planted in the data
- 2,000 entities
- Signal strength: 0.50 (sigmoid slope multiplier)
n_new_rows=5000n_prior=1000
What the system did
- Training rows built: 2,000
- Features used in model: 3
- Model selected:
DeploymentScoringRunner
What we checked — all assertions passed
| Status | Assertion | Detail |
|---|---|---|
| ✓ | deployment_completed_cleanly |
clean run |
| ✓ | scored_row_count_eq |
scored 5000 rows, expected 5000 |
| ✓ | prediction_distribution_in_band |
predictions in [0.000, 1.000] (mean 0.501) |
| ✓ | prior_predictions_unchanged |
prior predictions row count unchanged (1000) |
How this could have gone wrong
(no assertion descriptions provided)
Why this case matters
Backfills are common when an upstream pipeline catches up after an outage or schema change. The deployment must handle the big batch WITHOUT silently truncating or batching weirdly. 1 prediction per new row, full stop.
Reproducing
# from auto_insight_api/
python -m validation.v2 run dc_008 --pipeline deployment -v
- Case config:
validation/v2/cases/dc_008_high_volume.yaml - Data shape:
deployment_high_volume - Analytics type:
deployment