Run artifact · retrieval
retrieval-fixture
System fixture-predictions evaluated against synthetic-v1. This page presents only approved public evidence.
Verification: VERIFIEDData: SYNTHETIC_FIXTUREClaim: INTEGRATION_ONLY
Artifact ID
demo-retrieval-fixture-v1
Claim scope: Integration Only.
Aggregate values
Metrics
Hit Rate@5
0.6
k preserved in metric name · demo-retrieval-fixture-v1
MRR
0.6
k preserved in metric name · demo-retrieval-fixture-v1
nDCG@5
0.523
k preserved in metric name · demo-retrieval-fixture-v1
Precision@5
0.16
k preserved in metric name · demo-retrieval-fixture-v1
Recall@5
0.5
k preserved in metric name · demo-retrieval-fixture-v1
Failure summary
2 labeled failures- THQA-002SHOULD_ABSTAIN
- THQA-004RETRIEVAL_MISS
Provenance
- Dataset
- retrieval-fixture · synthetic-v1
- System
- fixture-predictions
- Evaluation
- retrieval
- Recorded
- 15 Jan 2026
- Top-k
- 5
- Evaluator
- retrieval: deterministic-metrics-v1
Record-level evidence
5 approved records| Record | Status | Metrics | Retrieved IDs | Relevant IDs |
|---|---|---|---|---|
| THQA-001 | Pass | Hit Rate@5 1 · MRR 1 · nDCG@5 1 · Precision@5 0.2 · Recall@5 1 | doc-th-001, doc-th-999 | doc-th-001 |
| THQA-002 | SHOULD_ABSTAIN | Hit Rate@5 0 · MRR 0 · nDCG@5 0 · Precision@5 0 · Recall@5 0 | doc-th-999 | Not included |
| THQA-003 | Pass | Hit Rate@5 1 · MRR 1 · nDCG@5 1 · Precision@5 0.4 · Recall@5 1 | doc-th-003, doc-th-002 | doc-th-002, doc-th-003 |
| THQA-004 | RETRIEVAL_MISS | Hit Rate@5 0 · MRR 0 · nDCG@5 0 · Precision@5 0 · Recall@5 0 | doc-th-999 | doc-th-004 |
| THQA-005 | Pass | Hit Rate@5 1 · MRR 1 · nDCG@5 0.613 · Precision@5 0.2 · Recall@5 0.5 | doc-th-005, doc-th-999 | doc-th-005, doc-th-006 |
Limitations
- • Metrics are deterministic fixture evidence with no external model or provider inference.
- • Synthetic retrieval fixture only; not a production workload or generalization claim.
Contract notes
Metric names and denominators are inherited from the deterministic evaluator output. No combined quality score, missing denominator, raw response, prompt, corpus text, or hidden reasoning is inferred here.
Schema public-evidence-v1 · Verification text remains visible alongside color.