500 Invented Support Tickets Followed from First Message to Final Fix, with Outcomes and Timestamps from 2023 to 2025

A support queue is a good picture of what company records look like when they hold a whole task: a customer writes in, an agent answers, something is tried, and the ticket ends with a result. This dataset imitates that shape with 500 made-up tickets, each one kept as a connected thread with the outcome recorded beside it and timestamps spread over three years, 2023 to 2025. Each ticket has an id, opening and closing times, hours to resolve, one of eight categories (double charge, login failure, export error, shipping delay, feature question, cancellation, slow performance, data correction), a channel (email, chat, web form or phone note), a priority, the product area, and the full list of messages with their timestamps and who wrote them (customer, agent or system). The resolution block records the outcome, the action taken, whether it was fixed on first contact, whether the ticket was reopened and the customer rating, which is sometimes missing, as in real queues. Good uses: testing classifiers for category, outcome or time to resolve; practising how to summarise a whole thread into its result; and showing a reviewer what connected, outcome-bearing records look like before any real data is discussed. Every person and company is a placeholder (Customer 1, Agent 3) and nothing refers to a real customer or product. Because the tickets come from templates they are more regular than real traffic, so treat it as a practice set. Files: train.jsonl and generate.py, which rebuilds the set from a fixed seed. A rough way to see how complete records feed into a data valuation: https://aitrainingdatabroker.com/data-value-calculator?utm_source=zenodo&utm_campaign=dataset-tickets