Run Statuses & Failures
What every Monitor Run status means and when capacity is consumed.
A Monitor Run moves through a fixed pipeline. The report keeps the failure reason when the pipeline cannot complete.
| Status | Meaning |
|---|---|
| Queued | Accepted and waiting for a worker |
| Ingesting | Freezing or preparing the corpus |
| Sampling | Selecting eligible records and critical-category coverage |
| Replaying | Generating candidate and, when requested, incumbent outputs |
| Evaluating | Running validators and the full judge panel |
| Aggregating | Applying the frozen decision contract |
| Completed | The receipt and recommendation are available |
| Failed | A blocking error stopped the Run |
| Cancelled | The Run was cancelled before completion |
Before inference starts
Preflight checks the exact production model, corpus readiness, observed control, sample and critical-category evidence, model access, hosted generation ceiling, and pooled allowance. A failure here does not consume a Run.
After inference starts
Once the first provider request begins, the Run counts against the allowance even if later provider or worker work fails. Existing replay and judgment records are idempotent, so an internal job retry does not duplicate completed calls.
What to do
Read the failure shown on the Run report and correct the Monitor, source, or Run configuration. Start a new Run only when the screen asks you to; do not assume a second submission will repair an invalid corpus or control boundary.