> For the complete documentation index, see [llms.txt](https://docs.unitlab.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.unitlab.ai/solutions/program-design-guides/quality-governance-playbook.md).

# Quality governance playbook

Catch annotation problems before release with review stages, automatic checks, consensus and benchmarks.

Quality in Unitlab is a set of checks you combine: a review stage, automatic validation, consensus between annotators, and benchmarks against gold labels. The **QA** page shows the result for every item.

<figure><img src="https://2337177515-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fi6jXAxtoN6UXTLAJFexq%2Fuploads%2Fv2MsFzxb1ZrX6PiLaUl2%2Fsol-qa-checks.webp?alt=media" alt="QA page with Quality checks tab: items needing attention, validation problems and geometry findings per item" width="800"><figcaption><p>Quality checks list the items with validation or geometry problems, so reviewers start where it matters.</p></figcaption></figure>

## Pick the checks

| Check          | Catches                                      | Where               |
| -------------- | -------------------------------------------- | ------------------- |
| Review stage   | Wrong class, missed objects, bad edges       | Workflow            |
| Quality checks | Missing required properties, broken geometry | QA → Quality checks |
| Consensus      | Disagreement between annotators              | QA → Consensus      |
| Benchmarks     | Distance from gold labels                    | QA → Benchmarks     |
| Issues         | Questions raised on a specific spot          | QA → Issues         |

## Run it

<figure><img src="https://2337177515-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fi6jXAxtoN6UXTLAJFexq%2Fuploads%2FnEF9KhFtIX7rkhYu8IxN%2Fsol-task-queue.webp?alt=media" alt="Task queue with Annotate, Review and Complete stages and tasks with their status" width="800"><figcaption><p>The queue shows how many tasks sit in each stage.</p></figcaption></figure>

1. Write the rules once in **Instructions** and make required properties required in the ontology, so the editor enforces them.
2. Add a Review stage. Reviewers reject with a reason; the task returns to the annotator.
3. Before a release, open **QA → Quality checks** and clear every item that needs attention.
4. For high-stakes data, add consensus or a benchmark with gold labels and track the scores.
