run_2026-09-30_device-troubleshooting
v3.4 · device-troubleshooting v3.0
Claude Sonnet 5.5 · bge-m3 · hybrid (BM25 + dense, RRF) · support-prompt v7 on device-troubleshooting v3.0
Pass rate
84.5%
Faithfulness
0.89
Answer correctness
0.86
Context precision
0.78
Context recall
0.83
Latency p95
4.5s
Failures by category
66 failed samples · largest share: unsupported
| Category | Share |
|---|---|
| Unsupported claims | 27.3% |
| Incorrect factual answers | 21.2% |
| Missing context | 18.2% |
| Retrieval mismatch | 18.2% |
| Instruction-following failures | 10.6% |
| Other | 4.5% |
Run details
- Dataset
- device-troubleshooting v3.0 · 426 items
- Generator
- Claude Sonnet 5.5
- Retriever
- bge-m3 · hybrid (BM25 + dense, RRF)
- Prompt
- support-prompt v7
- Judge
- LLM judge · Claude Opus 5.5 · rubric v3
- Started
- Sep 30, 13:10 UTC
- Duration
- 1h 12m
- Triggered by
- Owen Reid
- Tokens / request
- 3,420 in · 214 out
- devices
All metrics
Retrieval, generation and operational metrics
| Metric | Group | This run |
|---|---|---|
| Pass rateShare of samples the judge marked as passing all rubric checks. | Outcome | 84.5% |
| FaithfulnessShare of answer claims supported by the retrieved context. | Generation | 0.89 |
| Answer correctnessAgreement with the gold answer on facts, not style. | Generation | 0.86 |
| Instruction adherenceRespect for tone, format, length, language and policy constraints. | Generation | 0.92 |
| Context precisionRank-weighted precision of relevant chunks passed to the generator. | Retrieval | 0.78 |
| Context recallShare of gold-answer statements supported by the retrieved context. | Retrieval | 0.83 |
| Recall@5Share of gold passages found in the top 5 retrieved chunks. | Retrieval | 0.87 |
| MRRMean reciprocal rank of the first relevant chunk. | Retrieval | 0.73 |
| NDCG@10Graded ranking quality of the top 10, normalized by the ideal order. | Retrieval | 0.80 |
| Latency p50Median end-to-end time to final token. | Operational | 2.2s |
| Latency p9595th percentile end-to-end time to final token. | Operational | 4.5s |
| Cost per requestModel, embedding and reranking cost per answered request (USD). | Operational | $0.0046 |
Nightly trend
Nightly runs on the support-golden-v4 slice, Jul 11 to Oct 8. Dashed line marks this run.
- Pass rate · nightly
- This run · 84.5%
Across datasets
support-assistant v3.4 on every dataset in the project. Rows marked “matrix” come from the comparison matrix rather than a dedicated run.
| Dataset | Items | Pass rate | Faith. | Ctx recall | Recall@5 | p95 | Source |
|---|---|---|---|---|---|---|---|
| support-golden-v4 | 1,240 | 87.2% | 0.92 | 0.85 | 0.88 | 4.3s | Run · Oct 8 |
| billing-edge-cases | 312 | 81.7% | 0.87 | 0.82 | 0.86 | 4.0s | Run · Oct 2 |
| roaming-and-travel | 188 | 80.9% | 0.88 | 0.79 | 0.83 | 3.9s | Run · Sep 28 |
| device-troubleshootingthis run | 426 | 84.5% | 0.89 | 0.83 | 0.87 | 4.5s | This run |
| account-security | 154 | 88.3% | 0.92 | 0.85 | 0.89 | 3.7s | Run · Sep 25 |
| locale-formatting | 96 | 80.2% | 0.86 | 0.80 | 0.84 | 4.1s | Run · Oct 4 |
| synthetic-hard-negatives | 540 | 76.5% | 0.86 | 0.73 | 0.77 | 4.0s | matrix |
Traced samples
6 samples with full traces (3 failing, 3 passing), stratified by domain. Open a row to inspect retrieval and the judge's reasoning.
Is VoLTE supported on the Pixel 7a? Device troubleshooting FailMissingjudge 0.30 |
Hotspot doesn't work on my prepaid line. Device troubleshooting FailUnsupportedjudge 0.39 |
Personal hotspot is grayed out on my iPhone. Device troubleshooting Passjudge 0.90 |
How do I move my eSIM to my new phone? eSIM & activation Passjudge 0.87 |
How do I turn on VoLTE on my Android phone? Device troubleshooting Passjudge 0.96 |
Wi-Fi calling isn't working on my Samsung. Device troubleshooting FailMissingjudge 0.55 |
6 rows