> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cortex-labs.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Run ingestion and inspect processing
> Start bounded collection, inspect processing gaps and recover only the work that needs another attempt.
Use this procedure after choosing a supported source and destination dataset. You need dataset ingestion authority and permission to run the job.
## Check the job before starting
1. Open the dataset's job, or find it in **Forge → Jobs & Triggers**.
2. Confirm the source, destination, bundle version and saved parameters.
3. Inspect secret status. Ask an Admin to resolve **Not set** or **Revoked** prerequisites in Console.
4. Check date bounds, input limits, spend limits, retention and existing automatic enrichment.
5. For a custom Ingest bundle, confirm a successful test for this job and version.
System bundles have a separate publication review and do not require your own job test. A Format job reads webhook deliveries only. It offers neither **Run now** nor a test run.
## Start and follow collection
For a manual Ingest job, choose **Run now**, inspect any run overrides and submit once. Retain the run identifier.
A schedule or webhook creates an occurrence record. Inspect its action and resulting run separately. A repeated delivery identity joins the same occurrence.
Open the run. Check the principal, actual bundle version, input counts, collected material and extraction progress. **Parsing the last files** means collection ended while parsing continues.
| Outcome | Interpretation |
| - | - |
| Completed input | The requested input settled successfully. Inspect its content and downstream processing. |
| Skipped input | The recorded reason explains why no processing occurred. A skip can be an expected outcome. |
| Completed with errors | Some inputs failed. Successful inputs remain available. |
| Failed run | Collection failed, an input remained pending or no input completed. All-skipped runs are an exception. |
| No new data | The run can complete without new material. Check duplicates, scope and source availability. |
## Inspect processing separately
Open representative files and fragments. Check extracted text, page order, OCR and media locators against the retained source.
Confirm embedding and enrichment status independently. A successful collection run does not establish that every item is searchable or enriched. Check unsupported formats, unreadable pages, missing media and processing horizons.
New uploads to the same path create new file versions. Reprocessing can replace current derived content while retained work keeps its selected evidence versions.
## Resolve errors without repeating collection unnecessarily
Open the dataset **Errors** list. Read the reason before selecting an action.
| Action | Effect to check |
| - | - |
| Re-run a file | Parses retained run bytes with the last parse version, or reprocesses an uploaded file. |
| Re-run failures | Retries failed enrichment items and eligible file/run recovery independently. Inspect each result. |
| Re-run a run | Recovers retained input when available; otherwise starts another run with the same input. |
| Skip | Records that the failed item is intentionally excluded. It does not turn missing evidence into successful coverage. |
A file in a working run cannot be rerun concurrently. Wait for the run to settle. Do not reset a checkpoint to repair a parsing failure.
Finish by checking the original failed inputs, available coverage and new attempt records. Record remaining gaps in your investigation. For expiring input, see [Streaming data and retention](/vantage/bring-in-data/streaming-and-retention).
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.