Seeded Schema-Driven Test Data Generator

Flat CSV supports object records with scalar/null fields only. Nested objects, arrays and primitive roots are refused; CSV converts types to text and prefixes spreadsheet formula-like cells and leading-zero identifiers. Negative fixture supports one mutation only: remove one explicitly required root property from a validated object, then independently confirm schema validation fails. It is labelled invalid and never mixed into positive fixture exports.

Draft 2020-12 generation subset: single core types, properties/required, items, enum/const, numeric bounds, string/array lengths and local #/$defs references. Formats: UUID, Gregorian date and UTC date-time with Z. Patterns, composition, type unions and remote references block generation. Additional properties are never synthesized.

Priority: const → valid enum → enabled examples → enabled default → generated value. Invalid selected examples/defaults fail closed. Array size is clamped to schema bounds. One fixture exports its value; multiple fixtures export a JSON array, with each item validated separately.

Reproducibility: integronauts-fixtures-v1, NFC seed + fixture index, FNV-1a and Mulberry32. Same schema, seed, options and version produce identical JSON. Dates use a fixed 2020–2029 range. Limits: 64 KiB schema, 256 KiB output, 24 levels, 2,000 assessed schema nodes, 10,000 generated nodes, 200 generated arrays, 128 reference traversals, 50 items per array, 1,024 characters per string, 20 fixtures. Numbers use JavaScript precision; unsafe ranges are refused.

Schemas, imported files and generated fixtures stay in this browser. No network reference resolution, logs, URL payloads or automatic storage. Generated values are synthetic test data; seed changes vary values only where the schema allows it.