Customer Support Conversation Dataset (500 multi-turn dialogues)

500 original, entirely synthetic multi-turn customer-support dialogues (dialogues.jsonl), each set at one of 30 fictional companies or products — no real company, brand, or individual is represented anywhere in the dataset. Each record has a messages array (2-6 alternating turns), a scenario label, and a resolution status (resolved, escalated, or pending_info). Conversations span 15 scenario types — billing, subscription changes, troubleshooting, refunds, shipping, account access, general product questions, feature requests, complaints, positive feedback, escalation to a human, follow-up on a prior ticket, cancellation, upgrade/downgrade, and onboarding — varying customer tone, resolution path, and the product involved so records read as distinct interactions rather than templated duplicates; all 500 opening messages are unique. Verified programmatically to contain no real personal data, phone numbers, addresses, payment-card numbers, API keys, or shell commands, and only 9 reused @example.com placeholder addresses appear anywhere in the file. Intended for supervised fine-tuning, tone/style conditioning, training a resolution-routing classifier, or as a stratified evaluation set for a support assistant.