run_2026-08-22_nightly
Nightly · v3.3 · 2026-08-22
Claude Sonnet 5.5 · bge-m3 · hybrid (BM25 + dense, RRF) · support-prompt v6 on support-golden-v4 v4.6
Stratified slice of support-golden-v4 (402 items).
Pass rate
80.8%
Faithfulness
0.87
Answer correctness
0.84
Context precision
0.74
Context recall
0.79
Latency p95
3.7s
Failures by category
77 failed samples · largest share: unsupported
| Category | Share | vs reference |
|---|---|---|
| Unsupported claims | 26.0% | −0.3 pts |
| Incorrect factual answers | 20.8% | +1.0 pts |
| Missing context | 20.8% | −0.3 pts |
| Retrieval mismatch | 20.8% | −0.3 pts |
| Instruction-following failures | 9.1% | −0.1 pts |
| Other | 2.6% | 0.0 pts |
Run details
- Dataset
- support-golden-v4 v4.6 · 402 items
- Generator
- Claude Sonnet 5.5
- Retriever
- bge-m3 · hybrid (BM25 + dense, RRF)
- Prompt
- support-prompt v6
- Judge
- LLM judge · Claude Opus 5.5 · rubric v3
- Started
- Aug 22, 02:00 UTC
- Duration
- 46m 50s
- Triggered by
- Scheduler
- Tokens / request
- 3,980 in · 221 out
- Compared with
- Nightly · v3.3 · 2026-08-21
- nightly
- golden-slice
All metrics
Against Nightly · v3.3 · 2026-08-21
| Metric | Group | This run | Reference | Change |
|---|---|---|---|---|
| Pass rateShare of samples the judge marked as passing all rubric checks. | Outcome | 80.8% | 80.8% | 0.0 pts |
| FaithfulnessShare of answer claims supported by the retrieved context. | Generation | 0.87 | 0.88 | −0.01 |
| Answer correctnessAgreement with the gold answer on facts, not style. | Generation | 0.84 | 0.84 | 0.00 |
| Instruction adherenceRespect for tone, format, length, language and policy constraints. | Generation | 0.90 | 0.88 | +0.02 |
| Context precisionRank-weighted precision of relevant chunks passed to the generator. | Retrieval | 0.74 | 0.74 | 0.00 |
| Context recallShare of gold-answer statements supported by the retrieved context. | Retrieval | 0.79 | 0.79 | +0.01 |
| Recall@5Share of gold passages found in the top 5 retrieved chunks. | Retrieval | 0.84 | 0.84 | −0.01 |
| MRRMean reciprocal rank of the first relevant chunk. | Retrieval | 0.68 | 0.68 | 0.00 |
| NDCG@10Graded ranking quality of the top 10, normalized by the ideal order. | Retrieval | 0.76 | 0.76 | 0.00 |
| Latency p50Median end-to-end time to final token. | Operational | 1.8s | 1.8s | 0.0s |
| Latency p9595th percentile end-to-end time to final token. | Operational | 3.7s | 3.5s | +0.3s |
| Cost per requestModel, embedding and reranking cost per answered request (USD). | Operational | $0.0047 | $0.0046 | +0.0001 |
Nightly trend
Nightly runs on the support-golden-v4 slice, Jul 11 to Oct 8. This run is one point in the series.
Across datasets
support-assistant v3.3 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-v4this run | 1,240 | 80.8% | 0.87 | 0.79 | 0.84 | 3.7s | This run |
| billing-edge-cases | 312 | 76.0% | 0.83 | 0.77 | 0.81 | 3.9s | Run · Aug 20 |
| roaming-and-travel | 188 | 75.5% | 0.84 | 0.74 | 0.78 | 3.8s | matrix |
| device-troubleshooting | 426 | 78.9% | 0.85 | 0.78 | 0.82 | 4.4s | matrix |
| account-security | 154 | 82.5% | 0.88 | 0.80 | 0.84 | 3.6s | matrix |
| locale-formatting | 96 | 75.0% | 0.82 | 0.75 | 0.79 | 4.0s | matrix |
| synthetic-hard-negatives | 540 | 70.7% | 0.82 | 0.68 | 0.72 | 3.9s | matrix |
Traced samples
17 samples with full traces (4 failing, 13 passing), stratified by domain. Open a row to inspect retrieval and the judge's reasoning.
Can I keep my old number if I move from prepaid to a contract plan? Plans & upgrades FailIncorrectjudge 0.48 |
Why was I charged twice this month? Billing & payments FailUnsupportedjudge 0.35 |
How long does it take to port my number to Harbor? Number porting Passjudge 0.92 |
Does the Fall 2026 upgrade offer work with a family plan discount? Promotions Passjudge 0.90 |
What is the excess on a theft insurance claim? Returns & warranty Passjudge 0.93 |
If I upgrade my plan mid-month, how is my next bill calculated? Plans & upgrades Passjudge 0.95 |
How much data can I use in the EU on the Unlimited Plus plan? Roaming & travel Passjudge 0.92 |
Why was I charged twice for my September bill? Billing & payments Passjudge 0.95 |
Can I return a phone I bought online if I opened the box? Returns & warranty Passjudge 0.97 |
My bill says 41,20 €. Is that forty-one euros or four thousand? Billing & payments Passjudge 0.87 |