run_2026-09-25_account-security
v3.4 · account-security v1.2
Claude Sonnet 5.5 · bge-m3 · hybrid (BM25 + dense, RRF) · support-prompt v7 on account-security v1.2
Pass rate
88.3%
Faithfulness
0.92
Answer correctness
0.89
Context precision
0.80
Context recall
0.85
Latency p95
3.7s
Failures by category
18 failed samples · largest share: unsupported
| Category | Share |
|---|---|
| Unsupported claims | 27.8% |
| Incorrect factual answers | 22.2% |
| Missing context | 16.7% |
| Retrieval mismatch | 16.7% |
| Instruction-following failures | 11.1% |
| Other | 5.6% |
Run details
- Dataset
- account-security v1.2 · 154 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 25, 15:20 UTC
- Duration
- 25m
- Triggered by
- Maya Collins
- Tokens / request
- 3,420 in · 214 out
- security
All metrics
Retrieval, generation and operational metrics
| Metric | Group | This run |
|---|---|---|
| Pass rateShare of samples the judge marked as passing all rubric checks. | Outcome | 88.3% |
| FaithfulnessShare of answer claims supported by the retrieved context. | Generation | 0.92 |
| Answer correctnessAgreement with the gold answer on facts, not style. | Generation | 0.89 |
| Instruction adherenceRespect for tone, format, length, language and policy constraints. | Generation | 0.94 |
| Context precisionRank-weighted precision of relevant chunks passed to the generator. | Retrieval | 0.80 |
| Context recallShare of gold-answer statements supported by the retrieved context. | Retrieval | 0.85 |
| Recall@5Share of gold passages found in the top 5 retrieved chunks. | Retrieval | 0.89 |
| MRRMean reciprocal rank of the first relevant chunk. | Retrieval | 0.75 |
| NDCG@10Graded ranking quality of the top 10, normalized by the ideal order. | Retrieval | 0.82 |
| Latency p50Median end-to-end time to final token. | Operational | 1.9s |
| Latency p9595th percentile end-to-end time to final token. | Operational | 3.7s |
| Cost per requestModel, embedding and reranking cost per answered request (USD). | Operational | $0.0042 |
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 · 88.3%
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-troubleshooting | 426 | 84.5% | 0.89 | 0.83 | 0.87 | 4.5s | Run · Sep 30 |
| account-securitythis run | 154 | 88.3% | 0.92 | 0.85 | 0.89 | 3.7s | This run |
| 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 (2 failing, 4 passing), stratified by domain. Open a row to inspect retrieval and the judge's reasoning.
How quickly is a line blocked after reporting an unauthorized SIM swap? Account & security Passjudge 0.90 |
Someone swapped my SIM without my permission. How fast can you block it? Account & security FailIncorrectjudge 0.08 |
Can the 2FA requirement be removed over chat if I lose my phone? Account & security Passjudge 0.91 |
I lost my phone with my 2FA app. How do I sign in to My Harbor? Account & security Passjudge 0.93 |
Someone called pretending to be Harbor and asked for my code. What do I do? Account & security Passjudge 0.91 |
Do I need to keep my old SIM active while porting my number in? Number porting FailMismatchjudge 0.03 |
6 rows