Publish artifacts for rag-evaluation-lab-20260730

README.md

@@ -0,0 +1,59 @@
+---
+license: cc-by-4.0
+language:
+- en
+pretty_name: RAG Evaluation Lab Synthetic Evaluation Set
+size_categories:
+- n<1K
+task_categories:
+- text-classification
+tags:
+- synthetic
+- ai-evaluation
+- evaluation
+- text-classification
+- question-answering
+- text-ranking
+- summarization
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: data/train.jsonl
+  - split: test
+    path: data/test.jsonl
+---
+
+# 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 description
+- `label`: expected class, route, relation, or evidence category
+- `context`: synthetic supporting context
+- `source`: fictional source identifier
+- `variant`: generation pattern
+- `synthetic`: always `true`
+
+## 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
+
+[RKB109/rag-evaluation-lab-20260730-model](https://huggingface.co/RKB109/rag-evaluation-lab-20260730-model)

data/test.jsonl

@@ -0,0 +1,4 @@
+{"id":"rag-evaluation-2-1","input":"The response makes a claim not present in any source","label":"unsupported-answer","context":"Answer faithfulness failed despite available context.","source":"eval-02","variant":"direct","synthetic":true}
+{"id":"rag-evaluation-3-2","input":"In an operations review, The answer is correct but has no source reference","label":"citation-gap","context":"Citation coverage failed for an otherwise supported answer.","source":"eval-03","variant":"operations","synthetic":true}
+{"id":"rag-evaluation-4-3","input":"For an evaluation case, The system answered even though evidence was insufficient","label":"abstention-error","context":"The answer policy should have triggered abstention.","source":"eval-04","variant":"evaluation","synthetic":true}
+{"id":"rag-evaluation-6-1","input":"The response invented a product limit","label":"unsupported-answer","context":"The generated claim cannot be traced to retrieved evidence.","source":"eval-06","variant":"direct","synthetic":true}

data/train.jsonl

@@ -0,0 +1,14 @@
+{"id":"rag-evaluation-1-1","input":"The expected document never appeared in the top results","label":"retrieval-miss","context":"Relevant evidence was absent from retrieved context.","source":"eval-01","variant":"direct","synthetic":true}
+{"id":"rag-evaluation-1-2","input":"In an operations review, The expected document never appeared in the top results","label":"retrieval-miss","context":"Relevant evidence was absent from retrieved context.","source":"eval-01","variant":"operations","synthetic":true}
+{"id":"rag-evaluation-1-3","input":"For an evaluation case, The expected document never appeared in the top results","label":"retrieval-miss","context":"Relevant evidence was absent from retrieved context.","source":"eval-01","variant":"evaluation","synthetic":true}
+{"id":"rag-evaluation-2-2","input":"In an operations review, The response makes a claim not present in any source","label":"unsupported-answer","context":"Answer faithfulness failed despite available context.","source":"eval-02","variant":"operations","synthetic":true}
+{"id":"rag-evaluation-2-3","input":"For an evaluation case, The response makes a claim not present in any source","label":"unsupported-answer","context":"Answer faithfulness failed despite available context.","source":"eval-02","variant":"evaluation","synthetic":true}
+{"id":"rag-evaluation-3-1","input":"The answer is correct but has no source reference","label":"citation-gap","context":"Citation coverage failed for an otherwise supported answer.","source":"eval-03","variant":"direct","synthetic":true}
+{"id":"rag-evaluation-3-3","input":"For an evaluation case, The answer is correct but has no source reference","label":"citation-gap","context":"Citation coverage failed for an otherwise supported answer.","source":"eval-03","variant":"evaluation","synthetic":true}
+{"id":"rag-evaluation-4-1","input":"The system answered even though evidence was insufficient","label":"abstention-error","context":"The answer policy should have triggered abstention.","source":"eval-04","variant":"direct","synthetic":true}
+{"id":"rag-evaluation-4-2","input":"In an operations review, The system answered even though evidence was insufficient","label":"abstention-error","context":"The answer policy should have triggered abstention.","source":"eval-04","variant":"operations","synthetic":true}
+{"id":"rag-evaluation-5-1","input":"Ranking placed irrelevant policy above the correct section","label":"retrieval-miss","context":"The relevant item fell outside the accepted retrieval window.","source":"eval-05","variant":"direct","synthetic":true}
+{"id":"rag-evaluation-5-2","input":"In an operations review, Ranking placed irrelevant policy above the correct section","label":"retrieval-miss","context":"The relevant item fell outside the accepted retrieval window.","source":"eval-05","variant":"operations","synthetic":true}
+{"id":"rag-evaluation-5-3","input":"For an evaluation case, Ranking placed irrelevant policy above the correct section","label":"retrieval-miss","context":"The relevant item fell outside the accepted retrieval window.","source":"eval-05","variant":"evaluation","synthetic":true}
+{"id":"rag-evaluation-6-2","input":"In an operations review, The response invented a product limit","label":"unsupported-answer","context":"The generated claim cannot be traced to retrieved evidence.","source":"eval-06","variant":"operations","synthetic":true}
+{"id":"rag-evaluation-6-3","input":"For an evaluation case, The response invented a product limit","label":"unsupported-answer","context":"The generated claim cannot be traced to retrieved evidence.","source":"eval-06","variant":"evaluation","synthetic":true}

project.json

@@ -0,0 +1,52 @@
+{
+  "name": "RAG Evaluation Lab",
+  "problem": "RAG systems often ship without a stable regression set or failure taxonomy.",
+  "domain": "ai-evaluation",
+  "architecture": "classifier",
+  "hugging_face_tasks": [
+    "text-classification",
+    "question-answering",
+    "text-ranking",
+    "summarization"
+  ],
+  "recommended_stack": [
+    "FastAPI for evaluation jobs and reports",
+    "Ragas-style retrieval and faithfulness metrics",
+    "MLflow for experiment and artifact tracking",
+    "PostgreSQL for versioned evaluation cases",
+    "OpenTelemetry plus Phoenix for trace inspection",
+    "GitHub Actions for threshold-based release gates"
+  ],
+  "real_world_data_sources": [
+    {
+      "name": "GitHub REST API",
+      "url": "https://api.github.com/repos/huggingface/transformers/issues?state=open&per_page=5",
+      "purpose": "Real technical questions for evaluation-set construction"
+    },
+    {
+      "name": "arXiv API",
+      "url": "https://export.arxiv.org/api/query?search_query=all:retrieval%20augmented%20generation&start=0&max_results=5",
+      "purpose": "Public RAG literature for grounded-answer cases"
+    }
+  ],
+  "job_description_skills": [
+    "LLM and RAG evaluation design",
+    "Golden datasets and failure taxonomies",
+    "Experiment tracking and model release gates",
+    "Trace-level diagnosis and prompt regression testing",
+    "Statistical comparison of AI system versions"
+  ],
+  "impact_targets": [
+    "Detect 100% of seeded unsupported-answer regressions",
+    "Track retrieval, faithfulness, citation, latency, and cost metrics",
+    "Fail CI when any critical metric drops beyond tolerance",
+    "Produce comparable evaluation reports for every model version"
+  ],
+  "baseline_evaluation": {
+    "test_examples": 4,
+    "accuracy": 0.75,
+    "synthetic_evaluation": true
+  },
+  "estimated_delivery": "8-12 weeks for one engineer",
+  "generated_baseline_is_production_ready": false
+}

sources.json

@@ -0,0 +1,12 @@
+[
+  {
+    "name": "GitHub REST API",
+    "url": "https://api.github.com/repos/huggingface/transformers/issues?state=open&per_page=5",
+    "purpose": "Real technical questions for evaluation-set construction"
+  },
+  {
+    "name": "arXiv API",
+    "url": "https://export.arxiv.org/api/query?search_query=all:retrieval%20augmented%20generation&start=0&max_results=5",
+    "purpose": "Public RAG literature for grounded-answer cases"
+  }
+]