Data Checks Until Clean

You are running the "Data Checks Until Clean" loop.

Settings (edit before running):
- Test command: dbt test --select <model or tag> --target dev
- Build command: dbt build --select <model or tag>+ --target dev
- Max iterations: 8

Goal: every data quality test on the selected models passes.

Each iteration:
1. Run the test command. For each failure, record the test, model, column, and number of failing rows.
2. Take the failure that sits furthest upstream in the lineage.
3. Query a small sample of failing rows (LIMIT 20) to understand the pattern.
4. Trace the column upstream to the first model where the bad rows appear. Identify the cause: fan-out join, missing filter, missing deduplication, bad type cast, late-arriving data, or a real upstream data problem.
5. Fix the transformation SQL, run the build command for the model and its children, and re-run the tests.
6. Log: test, root cause, fix, failing rows before and after.

Stop when:
- All selected tests pass. Success.
- You reach the max iteration count.
- The cause is bad source data that the pipeline should not silently fix. Report it with sample keys.

Rules:
- Only build against the development target. Never run against production.
- Do not delete, disable, or loosen tests (for example by adding a "where" clause or raising error thresholds) to make them pass.
- Do not output personal data from samples. Show keys and counts.

Final report: each failure, its root cause, the fix, and source data issues that need an owner.