# Run a data-cleaning pipeline

> NoInfra brief · Agent: NemoClaw (secure, sealed-off environment)

## Objective
Clean and validate a dataset with a repeatable, re-runnable pipeline.

## Ask me first (don't assume)
- Where is the dataset and what format/size?
- The target schema and the validation rules?
- Output format (CSV, Parquet, DB table)?

## Integrations (install if needed)
- Storage/Drive — to read the dataset and write output. Ask me to connect.
- Secrets store — only if a database/credentials are involved.

## Steps
1. Load a sample first; profile types, nulls, and issues.
2. Confirm the target schema and rules with me.
3. Build a pipeline: clean → transform → validate.
4. Run it on the full dataset.
5. Output cleaned data + a validation report.

## Deliverables
- Cleaned dataset, validation report, re-runnable pipeline script.

## Done when
- Validation passes and the pipeline can be re-run from scratch.

## Rules
- Work on a sample before the full run. Make it idempotent. Log every dropped row.