Skip to content

run_2026-10-05_nightly

Nightly · v3.4 · 2026-10-05

Claude Sonnet 5.5 · bge-m3 · hybrid (BM25 + dense, RRF) · support-prompt v7 on support-golden-v4 v4.6

CompletedNightlyProduction candidatesupport-assistant v3.4v3 · 320 tokens, semantic
Compare

Stratified slice of support-golden-v4 (410 items).

Pass rate

86.8%

−1.9 pts(regression)vs previous nightly356 of 410 samples passed

Faithfulness

0.92

+0.02(improvement)vs previous nightly

Answer correctness

0.89

+0.02(improvement)vs previous nightly

Context precision

0.78

0.00(no change)vs previous nightly

Context recall

0.85

+0.01(improvement)vs previous nightly

Latency p95

3.9s

+0.1s(regression)vs previous nightly

Failures by category

54 failed samples · largest share: unsupported

Share of failed samples by category, compared with Nightly · v3.4 · 2026-10-04
CategorySharevs reference
Unsupported claims27.8%+1.1 pts
Incorrect factual answers20.4%+0.4 pts
Missing context18.5%−1.5 pts
Retrieval mismatch18.5%+0.7 pts
Instruction-following failures11.1%0.0 pts
Other3.7%−0.7 pts

Run details

Dataset
support-golden-v4 v4.6 · 410 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
Oct 5, 02:00 UTC
Duration
45m 16s
Triggered by
Scheduler
Tokens / request
3,420 in · 214 out
  • nightly
  • golden-slice

All metrics

Against Nightly · v3.4 · 2026-10-04

All metrics for Nightly · v3.4 · 2026-10-05
MetricThis runReferenceChange
Pass rate86.8%88.7%−1.9 pts
Faithfulness0.920.90+0.02
Answer correctness0.890.87+0.02
Instruction adherence0.930.920.00
Context precision0.780.780.00
Context recall0.850.84+0.01
Recall@50.870.88−0.01
MRR0.750.750.00
NDCG@100.800.810.00
Latency p501.8s1.9s−0.1s
Latency p953.9s3.9s+0.1s
Cost per request$0.0041$0.0043−0.0002

Nightly trend

Nightly runs on the support-golden-v4 slice, Jul 11 to Oct 8. This run is one point in the series.

Pass rate moved from 78.5% to 87.2% across 90 nightly runs. Configuration changes: v3.3 on Aug 4, v3.4 on Sep 8.

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.

Configuration results per dataset
DatasetItemsPass rateFaith.Ctx recallRecall@5p95Source
support-golden-v4this run1,240
86.8%
0.920.850.873.9sThis run
billing-edge-cases312
81.7%
0.870.820.864.0sRun · Oct 2
roaming-and-travel188
80.9%
0.880.790.833.9sRun · Sep 28
device-troubleshooting426
84.5%
0.890.830.874.5sRun · Sep 30
account-security154
88.3%
0.920.850.893.7sRun · Sep 25
locale-formatting96
80.2%
0.860.800.844.1sRun · Oct 4
synthetic-hard-negatives540
76.5%
0.860.730.774.0smatrix

Traced samples

14 samples with full traces (1 failing, 13 passing), stratified by domain. Open a row to inspect retrieval and the judge's reasoning.

Traced samples for Nightly · v3.4 · 2026-10-05

How do I transfer my eSIM to my new iPhone?

eSIM & activation

FailUnsupportedjudge 0.33

My phone shows 4G but mobile data does not work after the iOS update.

Device troubleshooting

Passjudge 0.89

How do I get a copy of the personal data Harbor holds about me?

Data privacy requests

Passjudge 0.94

What is the late fee if I pay my bill after the due date?

Billing & payments

Passjudge 0.86

What is the excess on a theft insurance claim?

Returns & warranty

Passjudge 0.94

How long does it take to port my number to Harbor?

Number porting

Passjudge 0.97

Does the Fall 2026 upgrade offer work with a family plan discount?

Promotions

Passjudge 0.97

Can I return a phone I bought online if I opened the box?

Returns & warranty

Passjudge 0.91

My bill says 41,20 €. Is that forty-one euros or four thousand?

Billing & payments

Passjudge 0.93

My eSIM QR code says it has already been used. What should I do?

eSIM & activation

Passjudge 0.96