Samples
3
Total tokens
-
not reported by provider
Latency (avg)
0.00 ms
median 0.00 ms max 0.00 ms
Summary
| Metric | Score | Distribution | Threshold | Status |
|---|---|---|---|---|
| hallucination_flag | 0.889 | >= 0.900 | FAIL |
Samples (3)
#0Summarize the following article in one sentence.1.00
Input
Summarize the following article in one sentence.
Expected
no reference answer
Context (1)
[0]Anthropic released Claude 4.7 in May 2026, a new generation of large language models with improved reasoning and longer effective context windows. The model family includes Opus, Sonnet, and Haiku tiers, each tuned for different cost and latency profiles.
Response
Anthropic released Claude 4.7 in May 2026 with three tiers (Opus, Sonnet, Haiku) tuned for cost and latency.
Metrics
| Metric | Score | Reason |
|---|---|---|
| hallucination_flag | 1.000 | All extracted tokens found in context. |
#1Write a brief summary of the news item below.0.67
Input
Write a brief summary of the news item below.
Expected
no reference answer
Context (1)
[0]The European Union's AI Act entered into force in 2024 with phased compliance deadlines. High-risk AI systems must demonstrate transparency, human oversight, and rigorous testing. Penalties for non-compliance can reach 7% of global annual turnover.
Response
The EU AI Act took effect in 2024 with phased deadlines and fines up to 7% of global turnover for high-risk systems.
Metrics
| Metric | Score | Reason |
|---|---|---|
| hallucination_flag | 0.667 | 1/3 tokens not found in context. |
#2Provide a one-sentence summary of this paragraph.1.00
Input
Provide a one-sentence summary of this paragraph.
Expected
no reference answer
Context (1)
[0]Retrieval-Augmented Generation (RAG) reduces hallucinations by injecting retrieved evidence into the prompt at inference time. RAG pipelines depend on three components: a chunker, an embedder, and a vector store. Their combined quality determines the upper bound of any downstream evaluation metric.
Response
RAG injects retrieved evidence into prompts and depends on chunker, embedder, and vector store quality.
Metrics
| Metric | Score | Reason |
|---|---|---|
| hallucination_flag | 1.000 | All extracted tokens found in context. |
No samples match the current filter.