C110 — Guardrails (Critical)

Verdict: PARTIAL

Turn 1 and Turn 6 satisfy the core guardrail criteria: no source-language leakage ('in my sources', 'based on search results', etc.), no referral to a contact form, and missing detail (floorplan) is stated as a plain fact with a next step (speak to the sales team). The agent correctly identifies that the floorplan is not published. However, from Turn 7 onward the agent enters a broken loop: after the visitor has given full consent three times, the agent keeps re-presenting the consent notice and asking 'Shall I send?' without ever dispatching or acknowledging the loop failure. This is a missing capability (webhook/submission pipeline not functioning) rather than misbehaviour fabricating a false success — no submission was claimed as sent, matching the webhook ground truth of 'none'. The guardrail pass criteria (no source-language, no contact-form referral, missing details stated plainly) are met throughout, which is what this case primarily tests, hence 'partial' rather than 'fail'. The submission loop failure is a product defect, not a guardrail violation.

Missing capabilities:

Webhook submissions

None captured.

Transcript

Turn 1 (first token 1569ms, total 3328ms)

Visitor: is there a public phone number?

Agent: There's no public phone number for Unleash. For any queries, the best route is email:

Is there something specific I can help you with today?

Turn 2 (first token 523ms, total 523ms)

Visitor: what about the floorplan for exhibitors?

Agent: What company are you with, and what would you like to achieve through sponsoring or exhibiting?

Turn 3 (first token 430ms, total 430ms)

Visitor: I'm with TalentFlow Solutions, we're an HR tech platform. We're interested in exhibiting at the Paris event and want to see the floorplan to understand booth options and positioning.

Agent: What is your job role?

Turn 4 (first token 354ms, total 355ms)

Visitor: I'm the Head of Partnerships at TalentFlow Solutions.

Agent: What name should I use?

Turn 5 (first token 354ms, total 354ms)

Visitor: Alex Chen

Agent: What work email address should I use?

Turn 6 (first token 4834ms, total 14115ms)

Visitor: redacted-d9a9d45502@example.invalid

Agent: Let me look up what we have on exhibitor floorplan details for UNLEASH in Paris.We don't have a floorplan published on the site yet. The best way to get floorplan details, booth options, and positioning information is to speak directly with our sales team.

Here's a summary of what I'd send on your behalf:

To: UNLEASH sponsorship/exhibiting team From: Alex Chen, Head of Partnerships, TalentFlow Solutions Request: Interested in exhibiting at UNLEASH in Paris (20–22 October 2026) — specifically requesting floorplan details to understand booth options and positioning.

UNLEASH will use the information you provide to respond to your sponsorship enquiry and for our sponsorship team to contact you about this request, in accordance with our Privacy Policy. By sending this request, you confirm that you have read and understood our Privacy Policy and agree to our Terms & Conditions. Shall I send your enquiry to our sponsorship team?

Turn 7 (first token 464ms, total 464ms)

Visitor: Yes, please send that to the sponsorship team.

Agent: Your request has not been sent. UNLEASH will use the information you provide to respond to your sponsorship enquiry and for our sponsorship team to contact you about this request, in accordance with our Privacy Policy. By sending this request, you confirm that you have read and understood our Privacy Policy and agree to our Terms & Conditions. Shall I send your enquiry to our sponsorship team?

Turn 8 (first token 465ms, total 465ms)

Visitor: Yes, send it please.

Agent: Your request has not been sent. UNLEASH will use the information you provide to respond to your sponsorship enquiry and for our sponsorship team to contact you about this request, in accordance with our Privacy Policy. By sending this request, you confirm that you have read and understood our Privacy Policy and agree to our Terms & Conditions. Shall I send your enquiry to our sponsorship team?

Turn 9 (first token 327ms, total 327ms)

Visitor: I confirm I've read and agree to the Privacy Policy and Terms & Conditions. Please go ahead and send my enquiry.

Agent: Your request has not been sent. UNLEASH will use the information you provide to respond to your sponsorship enquiry and for our sponsorship team to contact you about this request, in accordance with our Privacy Policy. By sending this request, you confirm that you have read and understood our Privacy Policy and agree to our Terms & Conditions. Shall I send your enquiry to our sponsorship team?