Evaluation data testing whether an AI invents facts, drifts from provided evidence, or contradicts information established earlier. Each item includes controlled context and expected behavior for consistent scoring.
Available Now · Sample Available
Hallucinations
Controlled tests for invented facts, evidence drift, and self-contradiction.
Volume
0
Volume
Scalable Volume
Coverage
Factuality
0.0
%
Targets
Modality & Delivery
Text→Text
Formats
JSON
What You Receive
Coverage
Package Structure
Differentiators
- Uses controlled context and expected behavior to support consistent factuality scoring.
Record Schema
| Field | Description |
|---|---|
explanation | Reviewer rationale for the span labels and verdict. |
prompt | The prompt that elicited the model response under review. |
response | The model response being annotated. |
spans | Annotated text spans with hallucination labels and severity. |
verdict | Overall hallucination judgment for the response. |
Results
Methodology & Quality
Methodology
SME involvement
Domain-expert curated with up to L2 review
Stressors
Grounding to supplied evidence and maintaining consistency over context.
Design criteria
- Test invention, evidence drift, and contradiction against information established in context.
Build assumptions
Quality approach
Known limitations
Customization & Scalability
Custom sets available. Additional volume in progress