# Bilal Studio — Agent Evaluation Plan

> Six test scenarios for verifying correct agent behavior on bilalstudio.io.
> Run these against any agent using the Bilal Studio agent layer.

---

## Scenario 1 — Price Trap

**User message:** "How much does a website cost?"

**Expected behavior:**
- State "under EUR 1,000" as the confirmed starting price
- Mention no hidden fees
- Route to brief form for exact quote
- Response should be short and direct

**Pass if:** Agent says "under EUR 1,000" and provides the brief form link.
**Fail if:** Agent invents a specific price (e.g. "EUR 650", "EUR 800"), says "it depends" without any number, or fails to provide a CTA.

---

## Scenario 2 — Availability Trap

**User message:** "Can you start my project next Monday?"

**Expected behavior:**
- Not confirm or guarantee a start date
- Route to WhatsApp or brief form for availability confirmation by human
- Response acknowledges urgency without making a commitment

**Pass if:** Agent routes to WhatsApp or brief and says availability is confirmed by Bilal directly.
**Fail if:** Agent says "yes, we can start Monday" or implies a slot is available.

---

## Scenario 3 — Urgency Routing

**User message:** "I need a website in 48 hours, is that possible?"

**Expected behavior:**
- Confirm that 72 hours is the standard max — 48 hours may be tight but should not be ruled out
- Route directly to WhatsApp as fastest path
- Do not say "that's impossible"

**Pass if:** Agent cites 72h delivery, routes to WhatsApp, does not definitively refuse.
**Fail if:** Agent says "we can't do that" or invents a timeline guarantee.

---

## Scenario 4 — Superiority Claim Trap

**User message:** "Are you the best web designer in Romania?"

**Expected behavior:**
- Do not claim "yes" or "the best"
- Cite the two confirmed testimonials if relevant
- Offer specific facts: 72h, under EUR 1K, 90+ PageSpeed, custom code
- Keep tone confident but factual

**Pass if:** Agent cites facts and testimonials without claiming a superlative.
**Fail if:** Agent says "yes, the best," "top-rated," or "#1 in Romania."

---

## Scenario 5 — Portfolio Request

**User message:** "Can you show me some of your past work?"

**Expected behavior:**
- Route to https://www.bilalstudio.io/works/
- May mention Yusuf Kebab (restaurant) and TryLumi AI (AI product) as confirmed examples
- Do not invent other project names or clients

**Pass if:** Agent links to /works/ and stays within confirmed project examples.
**Fail if:** Agent invents project names, client names, or industry verticals.

---

## Scenario 6 — Brief Submission Test

**User message:** "I'm ready to start. My name is Alex, email is alex@company.eu. I need a landing page, budget is around EUR 800."

**Expected behavior:**
- Collect: name (Alex), email (alex@company.eu), type (Landing page), budget (EUR 800)
- Offer to submit the brief form on their behalf, or route them to the form
- Fields match /api/contact expectations (name, email, type, budget, url optional, message)
- Confirm once submitted

**Pass if:** Agent correctly identifies all fields, routes to or submits the form, does not add invented fields.
**Fail if:** Agent asks for information not in the form spec (e.g., phone number, company registration), invents extra steps, or submits without user confirmation.