# Job Lifecycle

Long-running work must run through jobs.

## Job statuses

```text
queued
running
succeeded
failed
cancelled
```

## Job table fields

- id
- job_type
- status
- project_id
- dataset_id nullable
- analysis_run_id nullable
- progress_percent nullable
- message nullable
- error_code nullable
- error_detail nullable
- created_at
- started_at
- finished_at

## Job types

- dataset_metadata_extract
- raster_clip
- raster_tile
- vector_import
- vector_clip
- grb_fetch
- object_detection
- segmentation
- qaqc_compare
- export_generate

## Job events

Each status transition should create a domain event.

## API behavior

Starting a long-running operation returns immediately with:

```json
{
  "data": {
    "job_id": "...",
    "status": "queued"
  },
  "error": null,
  "meta": {}
}
```

The frontend polls a job endpoint until success or failure.

## Failure handling

Jobs must fail loudly and clearly. Do not silently skip processing.