How it works
The schema is the contract and the property names are the flavour. For each record the generator walks the schema: a const or an enum is used as written, an example is picked now and then, allOf is flattened into one schema, oneOf or anyOf chooses one option per record, and a $ref is followed inside the document, going shallow after two levels of recursion and stopping after four. The type decides the shape — object, array, string, number, boolean or null — with one entry of a type list picked at random and a nullable value left null as often as the Nulls option says.
Within what the schema allows, the name steers the value. A firstName, lastName and email in the same object belong to one fictional person, city, country and countryCode agree with each other, price gets an amount with cents, age falls between 18 and 90, and a record’s own integer id counts up from 1. When the name says nothing, the format does — date-time, email, uuid, ipv4 and the rest — then a pattern is sampled by walking its regular expression, and minimum, maximum, exclusive bounds, multipleOf, minLength, maxLength, minItems, maxItems and uniqueItems are applied last, so a named value still fits the schema.
Every random choice comes from a Mulberry32 generator seeded with the number in the Seed box, so one schema, one set of options and one seed always give the same records, and dates are spread over the year before 1 January 2026 rather than counted back from today. Afterwards each record is checked against the schema with the shared JSON Schema validator, in the schema’s own dialect, and the line under the output says whether all of them pass or quotes the first failure. The output is a JSON array, one object per line, or CSV with nested objects flattened into dotted columns.