Evaluation

All submitted runs will be evaluated using the same Task 2 framework.

Evaluation will consider:

• validity classification;

• error-span detection;

• hallucination-type classification;

• correction quality;

• clinical appropriateness;

• avoidance of new hallucinations; and

• confidence calibration.

The main Task 2 results may include all valid submissions. Results will also be grouped by evidence-use category so that report-only, image-only, and multimodal systems can be analyzed appropriately.

If only a small number of runs are submitted in a category, those results may be presented as an experimental comparison rather than as a separate official ranking.