Skip to content

run_2026-08-16_nightly

Nightly · v3.3 · 2026-08-16

Claude Sonnet 5.5 · bge-m3 · hybrid (BM25 + dense, RRF) · support-prompt v6 on support-golden-v4 v4.6

CompletedNightlyIn productionsupport-assistant v3.3v2 · 512 tokens / 64 overlap
Compare

Stratified slice of support-golden-v4 (396 items).

Pass rate

81.3%

−1.0 pts(regression)vs previous nightly322 of 396 samples passed

Faithfulness

0.87

+0.01(improvement)vs previous nightly

Answer correctness

0.85

0.00(no change)vs previous nightly

Context precision

0.74

+0.01(improvement)vs previous nightly

Context recall

0.80

+0.01(improvement)vs previous nightly

Latency p95

3.5s

−0.4s(improvement)vs previous nightly

Failures by category

74 failed samples · largest share: unsupported

Share of failed samples by category, compared with Nightly · v3.3 · 2026-08-15
CategorySharevs reference
Unsupported claims25.7%−1.1 pts
Incorrect factual answers20.3%+0.6 pts
Missing context21.6%+0.5 pts
Retrieval mismatch20.3%−0.9 pts
Instruction-following failures9.5%+1.0 pts
Other2.7%−0.1 pts

Run details

Dataset
support-golden-v4 v4.6 · 396 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 16, 02:00 UTC
Duration
44m 13s
Triggered by
Scheduler
Tokens / request
3,980 in · 221 out
  • nightly
  • golden-slice

All metrics

Against Nightly · v3.3 · 2026-08-15

All metrics for Nightly · v3.3 · 2026-08-16
MetricThis runReferenceChange
Pass rate81.3%82.3%−1.0 pts
Faithfulness0.870.87+0.01
Answer correctness0.850.850.00
Instruction adherence0.900.900.00
Context precision0.740.73+0.01
Context recall0.800.79+0.01
Recall@50.820.83−0.01
MRR0.700.690.00
NDCG@100.770.76+0.01
Latency p501.8s1.8s−0.1s
Latency p953.5s3.9s−0.4s
Cost per request$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.

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 v3.3 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
81.3%
0.870.800.823.5sThis run
billing-edge-cases312
76.0%
0.830.770.813.9sRun · Aug 20
roaming-and-travel188
75.5%
0.840.740.783.8smatrix
device-troubleshooting426
78.9%
0.850.780.824.4smatrix
account-security154
82.5%
0.880.800.843.6smatrix
locale-formatting96
75.0%
0.820.750.794.0smatrix
synthetic-hard-negatives540
70.7%
0.820.680.723.9smatrix

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.

Traced samples for Nightly · v3.3 · 2026-08-16

What happens if I go over my fair-use limit on Unlimited at home?

Plans & upgrades

FailMissingjudge 0.07

What's included in my data access request?

Data privacy requests

FailUnsupportedjudge 0.29

Can I change the billing email after my contract starts?

Billing & payments

FailMismatchjudge 0.12

What is the excess on a theft insurance claim?

Returns & warranty

Passjudge 0.91

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

Billing & payments

Passjudge 0.92

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

Data privacy requests

Passjudge 0.91

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

Returns & warranty

Passjudge 0.94

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

Device troubleshooting

Passjudge 0.94

Is SIM swap protection turned on by default?

Account & security

Passjudge 0.94

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

Billing & payments

Passjudge 0.94