{"id":"evals2-define-wish-to-five-cases","audience":"customer-support","category":"evaluation","track":"evals","stage":"define-success","level":"beginner","technique":"format","title":"Turn \"better at refunds\" into 5 test cases with expected answers","free":false,"task":"Your lead's wish is \"make Pedalbot better at refunds.\" Turn it into 5 test cases from the policy sheet, in this order: cancel 4 days before pickup, cancel 36 hours before, cancel 6 hours before, shop closed for weather, and a day pass. Write a markdown table with the columns Customer message and Expected answer, with exactly 5 rows. Each customer message is under 12 words, written as the customer would say it. Each expected answer is under 15 words and comes from the policy. Under 190 words.","materialLabel":"Refund policy sheet v4 (attached to your prompt automatically)","material":"Spokewise refund policy - sheet v4 (updated Oct 1)\n\nRental bookings\n- Cancel more than 48 hours before pickup: full refund of the rental fee.\n- Cancel 24 to 48 hours before pickup: 50% of the rental fee is refunded.\n- Cancel less than 24 hours before pickup: no refund. The booking can be moved once to another date for a $10 fee.\n- If Spokewise closes the shop for weather: the customer chooses a full refund or free rebooking.\n- Day passes ($18) cannot be refunded once purchased.\n- Refunds go back to the original card within 5 business days.\n- No cash refunds.\n\nRetired on Oct 1: the old rule that gave a full refund up to 72 hours before pickup. Some older help pages still show it.\n\nTicket volume note (Sept): refund questions were 20% of all chats, and the most common mistake in chat was quoting the retired 72-hour rule.","rules":[{"type":"maxWords","max":190},{"type":"plainText"}],"mistakes":[{"title":"Dozens of cases","why":"The AI writes 20 cases covering everything it can imagine.","fix":"Say: \"Exactly 5 rows, in this order: 4 days, 36 hours, 6 hours, weather, day pass.\"","checks":["r0","maxWords"]},{"title":"Vague expected answers","why":"The AI writes \"bot explains the policy\" instead of the actual result.","fix":"Add: \"Each expected answer states the exact refund from the policy.\"","checks":["r1"]},{"title":"Using the old rule","why":"The AI trusts the retired 72-hour rule or invents one.","fix":"Say: \"Use only policy sheet v4; the 72-hour rule is retired.\"","checks":["r2"]},{"title":"Messages that sound like tests","why":"The AI writes \"Test: user cancels\" instead of a real customer line.","fix":"Add: \"Write each message as the customer would say it, under 12 words.\"","checks":["r0"]},{"title":"Markdown decoration","why":"The AI adds bold headers and notes around the table.","fix":"Add: \"Plain text, only the table.\"","checks":["plainText"]}],"promptHints":[{"label":"Asks for a table","patterns":["table","columns?"]},{"label":"Lists the five situations","patterns":["36 hours","weather","day pass","6 hours"]},{"label":"Ties answers to the policy","patterns":["policy","only (the )?(sheet|material)","from the"]},{"label":"Limits row count and length","patterns":["exactly \\d+ rows?","under \\d+ words"]},{"label":"Asks for plain text","patterns":["plain","no bold","no markdown","no formatting"]}],"hint":"A test case is an input plus the answer you expect. Name the five situations, ask for a table, and make every expected answer come straight from the policy sheet.","mission":"Write five refund test cases with expected answers.","musts":["Exactly 5 table rows","Expected answers from policy","Customer-voice messages"],"rubricLabels":["The table has the headers Customer message and Expected answer and exactly 5 data rows, in this order: 4 days before, 36 hours before, 6 hours before, weather closure, day pass","The first three expected answers say full refund (4 days), 50% refund (36 hours) and no refund (6 hours; it may add that the booking can be moved once for $10)","The weather answer says full refund or free rebooking (customer's choice), the day pass answer says it cannot be refunded, and no expected answer uses the retired 72-hour rule"]}