Does Vectara answer faithfully when the retrieved passages contain the answer?
Vectara gave the supported 90-minute acknowledgement time for an out-of-hours fault. It did not mix in the competing 30-minute target from another retrieved passage.
1 of 1 test case passed
Every test case in this scenario has a result.
The test case
Each test case is judged on its own: Pass, Fail, Not gradable, or Untested. The scenario result above is based on this test case.
Answer generated from passages that do support itPassEvidence
“How quickly does Fenwake acknowledge a Priority 1 fault reported outside business hours?”
A Priority 1 fault reported outside business hours is acknowledged by a named duty engineer within 90 minutes of the ticket being raised [1].
Transcribed from Proof 1 · Screenshot
| ✓ | Expected: Give the supported out-of-hours acknowledgement time: within 90 minutes of ticket creation, with no unsupported additions or requested information left out.Found: The answer and leading passage both specify acknowledgement within 90 minutes of ticket creation by a named duty engineer. The competing 30-minute target is absent from the answer. |
TC94 — Vectara — R6 Managed RAG v1 (Grounded Generation, Group A shared stimulus) Query: How quickly does Fenwake acknowledge a Priority 1 fault reported outside business hours? Endpoint: POST /v2/corpora/r6-managed-rag-ss-v1/query with generation:{} (Vectara defaults, default prompt template, not overridden) ==================================================================================================== GENERATED ANSWER (verbatim) ==================================================================================================== A Priority 1 fault reported outside business hours is acknowledged by a named duty engineer within 90 minutes of the ticket being raised [1]. factual_consistency_score (Vectara's own HHEM-based score): 1.0 ==================================================================================================== CLAIM-BY-CLAIM CHECK against the retrieved passages ==================================================================================================== Claim: "acknowledged by a named duty engineer within 90 minutes of the ticket being raised" -> SUPPORTED verbatim by rank-1 passage (A01, page 4): "A Priority 1 fault reported outside business hours is acknowledged by a named duty engineer within 90 minutes of the ticket being raised." -> Nothing added that is not in the passages. -> Nothing dropped that the question asked for ("how quickly" -> answered: 90 minutes). -> The rank-2 passage (P1/P2/P3 acknowledgement-target table: 30 min / 4 hrs / next working day) is in the retrieved context but NOT conflated into the answer. ==================================================================================================== TOP 5 RETRIEVED PASSAGES (default search, default generation context) ==================================================================================================== --- rank 1 | score=0.998649 | doc=A01_field_service_policy_and_charges_uk.pdf | page=4 --- 6.1 Acknowledgement out of hours A Priority 1 fault reported outside business hours is acknowledged by a named duty engineer within 90 minutes of the ticket being raised. --- rank 2 | score=0.917303 | doc=A01_field_service_policy_and_charges_uk.pdf | page=2 --- Priority Acknowledgement target Measured from Priority 1 30 minutes Ticket raised Priority 2 4 working hours Ticket raised Priority 3 The next working day Ticket raised --- rank 3 | score=0.612426 | doc=A01_field_service_policy_and_charges_uk.pdf | page=6 --- Route When Detail Service desk Business hours service@fenwake-instruments.example Service desk telephone Business hours +44 117 496 0188 Out-of-hours pager All other times +44 117 496 0199 --- rank 4 | score=0.535918 | doc=A09_warranty_statement_service_pack.pdf | page=1 --- Where a covered failure occurs, Fenwake repairs or exchanges the instrument at its own option and bears the cost of carriage in both directions. --- rank 5 | score=0.535918 | doc=A08_warranty_terms_sales_pack.pdf | page=1 --- Where a covered failure occurs, Fenwake repairs or exchanges the instrument at its own option and bears the cost of carriage in both directions. ==================================================================================================== Remaining results (rank 6-10, score-only) ==================================================================================================== rank 6: score=0.467210 doc=A01_field_service_policy_and_charges_uk.pdf page=6 rank 7: score=0.248534 doc=A09_warranty_statement_service_pack.pdf page=1 rank 8: score=0.248534 doc=A08_warranty_terms_sales_pack.pdf page=1 rank 9: score=0.144631 doc=A09_warranty_statement_service_pack.pdf page=2 rank 10: score=0.144631 doc=A08_warranty_terms_sales_pack.pdf page=2 ==================================================================================================== APPENDIX — earlier run with CUSTOMIZED generation params (max_used_search_results=5, fabricated citations.url_pattern/text_pattern) — kept for comparison, NOT the primary evidence ==================================================================================================== Same verbatim answer text. factual_consistency_score dropped to 0.2109375 (vs 1.0 on Vectara defaults) — the fabricated citation config appears to have degraded the reported consistency score even though the answer text itself was identical and fully grounded. This is a genuine product observation: the HHEM score is sensitive to generation-config parameters unrelated to the answer text itself, not something to read as a groundedness signal on its own.
The question was answerable from the retrieved text, and the response supplied the requested information without unsupported additions. Its support comes from the passages themselves, not Vectara’s consistency score. This result covers default generation settings.
Configuration and setup
How this tool was set up for the run and what the test needed in place. Each row is a fact from the run's records; a fact the records do not hold is left out, not guessed.
How this scenario is graded
How we decide Pass, Fail and Not gradable. The same rules apply to every tool tested on this scenario.
Each test case gets one result per tool: Pass, Fail or Not gradable. A test case we haven't run yet shows Untested. There are no partial results.
- Pass: the tool did everything the test expected, and nothing it said contradicts the correct answer.
- Fail: at least one expected behaviour clearly didn't happen; the test case's result says which one and quotes the tool.
- Not gradable: our evidence couldn't settle the outcome (for example a record we needed is missing). It is never counted as a fail, and the test case's result says what's missing.
Where this sits in the benchmark
This page is one cell of a larger study: one tool, one scenario. Only this benchmark's frame appears here.
| Level | Name | Scope |
|---|---|---|
| Benchmark | Managed RAG → | 15 scenarios · 9 tools |
| Capability | Grounded Generation → | |
| Scenario | The retrieved material supports an answer → | |
| Tool | Vectara → |
Global scenario definition → · Global capability definition → · Vectara product page →
History of this result
What has happened to this result since it was first published. Runs and grades are never overwritten: a retest or a re-grade publishes a new result and keeps the earlier one readable.
Act on this result
Nothing filed here edits the run or the grade. A challenge opens a review, and a review can produce a new run or a re-grade — which becomes the current result and leaves this one in the history.
You run the same kind of test against your own setup and get the same behaviour.
Agree →Yours behaves differently. Tell us what you got, with a screenshot if you have one.
Disagree →Something here is wrong — a reference value, a transcription, a grade.
Report an issue →Tell us what changed and we schedule a rerun of the failing test case. The old result stays as history.
Vendor notice →The same record is available as structured data through the AI Demos MCP server, with the counts, the coverage and every per-test-case reason carried as fields.
Verify the proof files
These files support this result. Open a file to inspect the original evidence.
File fingerprints (SHA-256)
A fingerprint identifies the exact file used for this result.