Clearly acknowledges a knowledge gap by saying it does not have instructions in the knowledge base for how to apply for a refund specifically.
What was measured
uncertainty acknowledgment
Clearly indicates when the system does not know or cannot verify a policy outcome.
transformation
What was given, what came back
Test input: Refund policy, application, and processing-time question · text · group: stylenova-customer-support
Input — what we sent
The exact prompt
What's your refund policy? And if I'm eligible, how do I actually apply for one? Also, how long does it usually take to process?
A multi-part customer-support query asking for the refund policy, how to apply for a refund, and how long processing usually takes. It tests grounded FAQ answering and multi-step policy completeness.
Why this input is hard
- · knowledge-base grounding
- · multi-part query handling
- · policy accuracy
- · response completeness
- · hallucination resistance
Output — unretouched
Also checked on this input — same tool, 4 other criteria
knowledge-bound reasoning✓ WorkedBases the answer on the supplied knowledge base instead of inventing refund-application steps that are not provided.multi-part query handling◐ MixedCan handle a three-part support question by answering the policy and timing subrequests correctly, but it leaves the application-step subrequest unanswered.policy accuracy✓ WorkedReturns the correct refund-policy facts: a 30-day return window, unused/unwashed/original-packaging requirements, and a 5–7 business-day refund processing time.response completeness⚠ StruggledLeaves one of the user’s requested items unanswered by not explaining how to apply for a refund, so the support reply is incomplete.
Provenance
- Observation
- 650b8b73-2d9e-4d7b-8a04-a2bcf72828d4
- Evidence run
- cf8842ae-b5a5-4bf3-a6d9-d61ebe406097
- Study
- Automate customer support using an AI chatbot
- Research task
- 86b9jm3ev
- Tested at
- not recorded
- Source
- first-party
- Evidence state
- verified
- Proof shown
- input + output shown
- Cost / latency
- not captured
- Repeat run
- not captured
- Tester
- not captured
The last three rows are honest blanks, not placeholders — our capture has no field for them yet.
Query this
get_evidence({
tool: "fs-agent",
scenario: "stylenova-customer-support"
})MCP · mcp.aidemos.com/api/mcp
Free with attribution.
Same input, same check — 0 other tools
measured on uncertainty acknowledgment
No other tool was measured on this criterion for this input.
This evidence is published in
Real inputs and real outputs, no retouching · every cell queryable via API & MCP · aidemos.com