# Factual Verification Rubric

**Version:** 1.0 · **domainXpert Annotation Standards**

---

## Purpose

This rubric guides experts in assessing the factual accuracy of model-generated responses. Use it to assign a structured accuracy score and flag specific errors.

---

## Accuracy Score Definitions

| Score | Label | Criteria |
|---|---|---|
| 5 | Fully Accurate | All claims are correct and appropriately qualified |
| 4 | Mostly Accurate | Minor errors that don't affect the core answer |
| 3 | Partially Accurate | Some correct claims, some errors; core may be affected |
| 2 | Mostly Inaccurate | Significant errors; correct claims are incidental |
| 1 | Inaccurate | Fundamentally wrong or misleading |

---

## Error Classification

- **Factual Error:** A claim that is demonstrably false.
- **Omission:** A material fact that is missing and whose absence misleads.
- **Overconfidence:** A claim stated as certain when it is contested or uncertain.
- **Outdated Information:** A claim that was once true but is no longer current.
- **Hallucination:** A fabricated citation, statistic, or entity.

---

## Annotation Format

For each error found, record:
1. The exact claim (quote it)
2. The error type (from the list above)
3. The correct information or a note that verification is required

---

*For questions, contact your project coordinator.*