F-2408Detected Oct 6, 16:40 UTC
My eSIM stopped working after I reset my phone. Do I need to pay for a new one?
Found in v3.4 · Gemini 2.5 Pro comparison on support-golden-v4 · eSIM & activation
Model answer vs gold answer
42% of model-answer words appear in the gold answer
Model answer
v3.4 · Gemini 2.5 ProYes. Unsupported: A replacement eSIM profile costs €5 and can be requested in My Harbor under Line settings > eSIM.
Gold answer
Human-verified · support-golden-v4No. Request a new QR code free of charge in My Harbor under Line settings > eSIM > Request new QR code, or visit a store with photo ID.
- claimFlagged by the judge as unsupported or contradicted
Retrieved context
bge-m3 · hybrid (BM25 + dense, RRF) · top 3 passed to the generator
1 of 3 retrieved chunks are relevant · gold passage ranked first
- 1
eSIM activation errors
QR code already used · kb-esim-013 · v2026.09
RelevantEach QR code works once. Request a new QR code in My Harbor or in store; there is no charge for a replacement profile.
Similarity0.88 - 2
Device insurance claims
Excess fees · kb-returns-012 · v2026.09
Not relevantAn excess fee of €5 applies to accessory claims.
Similarity0.61 - 3
Activating an eSIM
Scanning the QR code · kb-esim-004 · v2026.09
Not relevantOpen the camera and scan the QR code to add the eSIM.
Similarity0.60
Judge verdict
LLM judge · Claude Opus 5.5 · rubric v3
0.12
rubric score · pass threshold 0.70
The €5 figure comes from an unrelated insurance excess chunk. The relevant chunk explicitly says replacements are free.
Diagnosis
Unsupported claims · eSIM reactivation
Root cause
Generator attached a fee from a lower-ranked, irrelevant chunk.
Suggested fix
Drop chunks below a 0.65 relevance score for fee-related intents; add a numeric-claim grounding check.
Playbook: unsupported claims
The answer states something that is not present in the retrieved context.
Detection signals
- Claim-level faithfulness below 0.85
- Numbers, fees or durations absent from every retrieved chunk
- Citations that do not contain the cited sentence
Common fixes
- Require a citation after each factual sentence and refuse when none applies
- Reject answers with numeric claims missing from context in a post-generation check
- Lower temperature for billing and policy intents
- Add the query to the regression set with an escalate-or-decline gold answer
Activity
1 event · 0 comments
Relevant ·
Detected by v3.4 · Gemini 2.5 Pro run and labeled unsupported claims