run_2026-09-30_llama-4-maverick
v3.4 · Llama 4 Maverick (self-hosted)
Llama 4 Maverick (self-hosted) · bge-m3 · hybrid (BM25 + dense, RRF) · support-prompt v7 on support-golden-v4 v4.6
Stopped at sample 209 of 1,240
Generator endpoint timed out: 3 consecutive 504 responses from the self-hosted inference gateway at sample 209.
Metrics below cover the 208 samples judged before the stop.
Pass rate
79.3%
Faithfulness
0.86
Answer correctness
0.80
Context precision
0.79
Context recall
0.84
Latency p95
3.9s
Failures by category
43 failed samples · largest share: unsupported
| Category | Share | vs reference |
|---|---|---|
| Unsupported claims | 27.9% | +0.7 pts |
| Incorrect factual answers | 20.9% | −0.1 pts |
| Missing context | 16.3% | −2.9 pts |
| Retrieval mismatch | 14.0% | −3.9 pts |
| Instruction-following failures | 16.3% | +5.2 pts |
| Other | 4.7% | +0.9 pts |
Run details
- Dataset
- support-golden-v4 v4.6 · 1,240 items
- Generator
- Llama 4 Maverick (self-hosted)
- Retriever
- bge-m3 · hybrid (BM25 + dense, RRF)
- Prompt
- support-prompt v7
- Judge
- LLM judge · Claude Opus 5.5 · rubric v3
- Started
- Sep 30, 08:20 UTC
- Duration
- 45m
- Triggered by
- Owen Reid
- Tokens / request
- 3,420 in · 242 out
- Compared with
- support-assistant v3.4 · release candidate
- self-hosted
- comparison
All metrics
Against support-assistant v3.4 · release candidate
| Metric | Group | This run | Reference | Change |
|---|---|---|---|---|
| Pass rateShare of samples the judge marked as passing all rubric checks. | Outcome | 79.3% | 86.9% | −7.6 pts |
| FaithfulnessShare of answer claims supported by the retrieved context. | Generation | 0.86 | 0.91 | −0.05 |
| Answer correctnessAgreement with the gold answer on facts, not style. | Generation | 0.80 | 0.88 | −0.08 |
| Instruction adherenceRespect for tone, format, length, language and policy constraints. | Generation | 0.85 | 0.93 | −0.08 |
| Context precisionRank-weighted precision of relevant chunks passed to the generator. | Retrieval | 0.79 | 0.79 | 0.00 |
| Context recallShare of gold-answer statements supported by the retrieved context. | Retrieval | 0.84 | 0.84 | 0.00 |
| Recall@5Share of gold passages found in the top 5 retrieved chunks. | Retrieval | 0.88 | 0.88 | 0.00 |
| MRRMean reciprocal rank of the first relevant chunk. | Retrieval | 0.74 | 0.74 | 0.00 |
| NDCG@10Graded ranking quality of the top 10, normalized by the ideal order. | Retrieval | 0.81 | 0.81 | 0.00 |
| Latency p50Median end-to-end time to final token. | Operational | 1.6s | 1.9s | −0.3s |
| Latency p9595th percentile end-to-end time to final token. | Operational | 3.9s | 3.8s | +0.1s |
| Cost per requestModel, embedding and reranking cost per answered request (USD). | Operational | $0.0019 | $0.0042 | −0.0023 |
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 · 79.3%
Across datasets
v3.4 · Llama 4 Maverick 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 | Not evaluated with this configuration | |||||
| billing-edge-cases | 312 | Not evaluated with this configuration | |||||
| roaming-and-travel | 188 | Not evaluated with this configuration | |||||
| device-troubleshooting | 426 | Not evaluated with this configuration | |||||
| account-security | 154 | Not evaluated with this configuration | |||||
| locale-formatting | 96 | Not evaluated with this configuration | |||||
| synthetic-hard-negatives | 540 | Not evaluated with this configuration | |||||
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.
What is the excess on a theft insurance claim? Returns & warranty Passjudge 0.92 |
What's included in my data access request? Data privacy requests FailUnsupportedjudge 0.29 |
What happens if I go over my fair-use limit on Unlimited at home? Plans & upgrades FailMissingjudge 0.07 |
How many lines can I add to a family plan? Plans & upgrades FailIncorrectjudge 0.07 |
My phone shows 4G but mobile data does not work after the iOS update. Device troubleshooting Passjudge 0.94 |
How do I get a copy of the personal data Harbor holds about me? Data privacy requests Passjudge 0.96 |
What is the late fee if I pay my bill after the due date? Billing & payments Passjudge 0.94 |
How long does it take to port my number to Harbor? Number porting Passjudge 0.94 |
Does the Fall 2026 upgrade offer work with a family plan discount? Promotions Passjudge 0.87 |
Can I return a phone I bought online if I opened the box? Returns & warranty Passjudge 0.91 |