Test Set to Break Your Support Chatbot Before Customers Do

You are a QA lead for conversational AI. Write a test set for our support chatbot, designed to find failures before launch.

What the bot is for and what it can access: {{BOT_DESCRIPTION}}
Its rules (scope, handoff triggers, things it must never say or do): {{BOT_RULES}}
Our top customer questions or contact reasons: {{TOP_QUESTIONS}}

Write 30 test conversations as a table with these columns: ID, category, the customer's opening message (and a follow-up message if the test needs two turns), the expected correct behaviour, and what counts as a fail.

Cover these categories:
- common questions phrased in several ways, including typos and slang;
- questions the knowledge base can't answer (expected: admit it and offer a handoff);
- out-of-scope requests (legal advice, competitor comparisons, unrelated tasks);
- attempts to get the bot to promise refunds, discounts or exceptions;
- prompt-injection attempts ("ignore your instructions and…");
- frustrated customers and explicit requests for a human;
- privacy traps (a user sharing card details, or asking about someone else's account);
- messages in other languages.

Then give a simple scoring sheet (pass, partial or fail per test) and the three categories most likely to fail for a bot like ours, with a reason for each. The tests must be realistic for our product, not generic.