RKB109/rag-evaluation-lab-20260720-model
Text Classification • Updated
id stringlengths 18 18 | input stringlengths 49 83 | label stringclasses 4
values | context stringclasses 6
values | source stringclasses 6
values | variant stringclasses 3
values | synthetic bool 1
class |
|---|---|---|---|---|---|---|
rag-evaluation-1-1 | The expected document never appeared in the top results | retrieval-miss | Relevant evidence was absent from retrieved context. | eval-01 | direct | true |
rag-evaluation-1-2 | In an operations review, The expected document never appeared in the top results | retrieval-miss | Relevant evidence was absent from retrieved context. | eval-01 | operations | true |
rag-evaluation-1-3 | For an evaluation case, The expected document never appeared in the top results | retrieval-miss | Relevant evidence was absent from retrieved context. | eval-01 | evaluation | true |
rag-evaluation-2-2 | In an operations review, The response makes a claim not present in any source | unsupported-answer | Answer faithfulness failed despite available context. | eval-02 | operations | true |
rag-evaluation-2-3 | For an evaluation case, The response makes a claim not present in any source | unsupported-answer | Answer faithfulness failed despite available context. | eval-02 | evaluation | true |
rag-evaluation-3-1 | The answer is correct but has no source reference | citation-gap | Citation coverage failed for an otherwise supported answer. | eval-03 | direct | true |
rag-evaluation-3-3 | For an evaluation case, The answer is correct but has no source reference | citation-gap | Citation coverage failed for an otherwise supported answer. | eval-03 | evaluation | true |
rag-evaluation-4-1 | The system answered even though evidence was insufficient | abstention-error | The answer policy should have triggered abstention. | eval-04 | direct | true |
rag-evaluation-4-2 | In an operations review, The system answered even though evidence was insufficient | abstention-error | The answer policy should have triggered abstention. | eval-04 | operations | true |
rag-evaluation-5-1 | Ranking placed irrelevant policy above the correct section | retrieval-miss | The relevant item fell outside the accepted retrieval window. | eval-05 | direct | true |
rag-evaluation-5-2 | In an operations review, Ranking placed irrelevant policy above the correct section | retrieval-miss | The relevant item fell outside the accepted retrieval window. | eval-05 | operations | true |
rag-evaluation-5-3 | For an evaluation case, Ranking placed irrelevant policy above the correct section | retrieval-miss | The relevant item fell outside the accepted retrieval window. | eval-05 | evaluation | true |
rag-evaluation-6-2 | In an operations review, The response invented a product limit | unsupported-answer | The generated claim cannot be traced to retrieved evidence. | eval-06 | operations | true |
rag-evaluation-6-3 | For an evaluation case, The response invented a product limit | unsupported-answer | The generated claim cannot be traced to retrieved evidence. | eval-06 | evaluation | true |
RAG Evaluation Lab Synthetic Dataset
Summary
This dataset contains 14 training examples and 4 held-out examples for RAG systems often ship without a stable regression set or failure taxonomy.
Every record is synthetic and includes:
input: query, event, or feature descriptionlabel: expected class, route, relation, or evidence categorycontext: synthetic supporting contextsource: fictional source identifiervariant: generation patternsynthetic: alwaystrue
Uses
- Reproducible unit and integration tests
- Baseline model training
- Evaluation harness development
- Schema and architecture demonstrations
Limitations
Synthetic cases validate the harness, not a production RAG system. Teams must add representative domain examples.
This dataset does not represent real users, patients, customers, production traffic, or licensed media. It must not be presented as real-world evidence.
Related Model
- Downloads last month
- 10
Models trained or fine-tuned on RKB109/rag-evaluation-lab-20260720-dataset