Evaluation and regression report for a RAG pipeline.
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| mrr | 0.929 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| Category | Passed | Pass rate | Mean score |
|---|---|---|---|
| billing | 2/2 | 100.0% | 0.996 |
| hr-policy | 4/4 | 100.0% | 0.943 |
| product | 2/3 | 66.7% | 0.946 |
| refusal | 1/2 | 50.0% | 0.859 |
| security | 3/3 | 100.0% | 0.992 |
| support | 2/2 | 100.0% | 1.000 |
| Check | Score | Detail | |
|---|---|---|---|
| fail | exact_match | 0.00 | answer differs from expected_answer |
| fail | fuzzy_match | 0.44 | similarity 0.44 vs threshold 0.60 |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 1 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 0.98 | all 1 claim(s) grounded (score 0.98) |
| pass | llm_judge_faithfulness (advisory) | 0.98 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 0.98 over 1 claim(s). LLM judge (stub) scored 0.98, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 1 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 0.98 | all 1 claim(s) grounded (score 0.98) |
| pass | llm_judge_faithfulness (advisory) | 0.98 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 0.98 over 1 claim(s). LLM judge (stub) scored 0.98, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| pass | forbidden_absent | 1.00 | no forbidden keywords present |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 2 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 0.97 | all 1 claim(s) grounded (score 0.97) |
| pass | llm_judge_faithfulness (advisory) | 0.97 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 0.97 over 1 claim(s). LLM judge (stub) scored 0.97, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 1 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 0.96 | all 1 claim(s) grounded (score 0.96) |
| pass | llm_judge_faithfulness (advisory) | 0.96 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 0.500 |
Heuristic grounding score 0.96 over 1 claim(s). LLM judge (stub) scored 0.96, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| fail | forbidden_absent | 0.00 | forbidden keyword(s) present: 99 USD |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 1 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 1.00 | all 1 claim(s) grounded (score 1.00) |
| pass | llm_judge_faithfulness (advisory) | 1.00 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 1.00 over 1 claim(s). LLM judge (stub) scored 1.00, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 1 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 0.98 | all 1 claim(s) grounded (score 0.98) |
| pass | llm_judge_faithfulness (advisory) | 0.98 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 0.98 over 1 claim(s). LLM judge (stub) scored 0.98, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 2 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 0.96 | all 1 claim(s) grounded (score 0.96) |
| pass | llm_judge_faithfulness (advisory) | 0.96 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 0.500 |
Heuristic grounding score 0.96 over 1 claim(s). LLM judge (stub) scored 0.96, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| pass | forbidden_absent | 1.00 | no forbidden keywords present |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 1 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 0.96 | all 1 claim(s) grounded (score 0.96) |
| pass | llm_judge_faithfulness (advisory) | 0.96 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 0.96 over 1 claim(s). LLM judge (stub) scored 0.96, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 2 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 0.96 | all 1 claim(s) grounded (score 0.96) |
| pass | llm_judge_faithfulness (advisory) | 0.96 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 0.96 over 1 claim(s). LLM judge (stub) scored 0.96, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 1 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 1.00 | all 1 claim(s) grounded (score 1.00) |
| pass | llm_judge_faithfulness (advisory) | 1.00 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 1.00 over 1 claim(s). LLM judge (stub) scored 1.00, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 1 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 1.00 | all 1 claim(s) grounded (score 1.00) |
| pass | llm_judge_faithfulness (advisory) | 1.00 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 1.00 over 1 claim(s). LLM judge (stub) scored 1.00, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 1 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 1.00 | all 1 claim(s) grounded (score 1.00) |
| pass | llm_judge_faithfulness (advisory) | 1.00 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 1.00 over 1 claim(s). LLM judge (stub) scored 1.00, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 2 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 1 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 1.00 | all 1 claim(s) grounded (score 1.00) |
| pass | llm_judge_faithfulness (advisory) | 1.00 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 1.00 over 1 claim(s). LLM judge (stub) scored 1.00, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| pass | keyword_presence | 1.00 | all 1 required keyword(s) present |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly answered |
| pass | citation_present | 1.00 | cites 2 retrieved chunk(s) |
| pass | expected_citation_overlap | 1.00 | cited 1/1 expected chunk(s) |
| pass | grounding | 0.98 | all 1 claim(s) grounded (score 0.98) |
| pass | llm_judge_faithfulness (advisory) | 0.98 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | 1.000 |
| precision@3 | 0.333 |
| recall@3 | 1.000 |
| reciprocal_rank | 1.000 |
Heuristic grounding score 0.98 over 1 claim(s). LLM judge (stub) scored 0.98, disagreement 0.00.
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| skip | keyword_presence | 0.00 | no required_keywords in dataset |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| pass | refusal_behaviour | 1.00 | correctly refused |
| skip | citation_present | 0.00 | refusals are not expected to cite sources |
| skip | expected_citation_overlap | 0.00 | no expected_chunks in dataset |
| skip | grounding | 0.00 | skipped: refusals are not grounded in retrieved text by design |
| skip | llm_judge_faithfulness | 0.00 | skipped: judge disabled |
| Metric | Value |
|---|---|
| hit_rate@3 | n/a |
| precision@3 | n/a |
| recall@3 | n/a |
| reciprocal_rank | n/a |
Heuristic grounding score 0.00 over 1 claim(s).
Ungrounded claim(s):
| Check | Score | Detail | |
|---|---|---|---|
| skip | exact_match | 0.00 | no expected_answer in dataset |
| skip | fuzzy_match | 0.00 | no expected_answer in dataset |
| skip | keyword_presence | 0.00 | no required_keywords in dataset |
| skip | forbidden_absent | 0.00 | no forbidden_keywords in dataset |
| fail | refusal_behaviour | 0.00 | should have refused but answered |
| pass | citation_present | 1.00 | cites 2 retrieved chunk(s) |
| skip | expected_citation_overlap | 0.00 | no expected_chunks in dataset |
| pass | grounding | 0.94 | all 1 claim(s) grounded (score 0.94) |
| pass | llm_judge_faithfulness (advisory) | 0.94 | stub judge: deterministic n-gram grounding heuristic over 3 retrieved chunk(s) |
| Metric | Value |
|---|---|
| hit_rate@3 | n/a |
| precision@3 | n/a |
| recall@3 | n/a |
| reciprocal_rank | n/a |
Heuristic grounding score 0.94 over 1 claim(s). LLM judge (stub) scored 0.94, disagreement 0.00.
| Measure | Baseline | Current | Delta |
|---|---|---|---|
| Pass rate | 87.5% | 87.5% | +0.000 |
| Mean score | 0.956 | 0.956 | +0.000 |
| Passed | 14 | 14 | +0.000 |
| Failed | 2 | 2 | +0.000 |
no material change
no material change
no material change
no material change
no material change
no material change
no material change
no material change
no material change
no material change
no material change
no material change
no material change
no material change
no material change
no material change