F-2409Detected Oct 2, 10:04 UTC
I upgraded from Essential to Unlimited on the 18th. Why is my bill €41.20 instead of €34.90?
Found in v3.4 · billing-edge-cases v2.1 on billing-edge-cases · Billing & payments
Model answer vs gold answer
29% of model-answer words appear in the gold answer
Model answer
support-assistant v3.4Your bill is higher because Unsupported: upgrades are charged a one-off €6.30 upgrade fee kb-plans-008.
Gold answer
Human-verified · billing-edge-casesUpgrades take effect immediately, so the bill includes a pro-rated charge for Unlimited from the 18th, a credit for unused Essential days, and the full Unlimited price for the next cycle. There is no upgrade fee.
- claimFlagged by the judge as unsupported or contradicted
Retrieved context
bge-m3 · hybrid (BM25 + dense, RRF) · top 3 passed to the generator
2 of 3 retrieved chunks are relevant · gold passage ranked first
- 1
Changing your plan
Upgrades · kb-plans-008 · v2026.09
RelevantUpgrades take effect immediately. Your next bill shows a pro-rated charge for the new plan and a credit for the old plan.
Similarity0.84 - 2
Understanding your monthly bill
Pro-rated charges · kb-billing-014 · v2026.09
RelevantPro-rated charges cover the days between a plan change and the end of the cycle.
Similarity0.80 - 3
Understanding your monthly bill
One-off charges · kb-billing-014 · v2026.09
Not relevantOne-off charges include roaming packs, device payments and late fees.
Similarity0.66
Judge verdict
LLM judge · Claude Opus 5.5 · rubric v3
0.06
rubric score · pass threshold 0.70
The answer invents an upgrade fee by subtracting the two amounts in the question. The retrieved context explains proration and contains no fee.
Diagnosis
Unsupported claims · Plan change proration
Root cause
Arithmetic on numbers from the question was presented as a policy fact.
Suggested fix
Add a rule to explain amounts only through retrieved bill components; add proration worked examples to the few-shot set.
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 support-assistant v3.4 run and labeled unsupported claims