generating-synthetic-test-datalisted

# Generating Synthetic Test Data ## When to use - Creating fixtures for pipeline/dbt unit and integration tests. - Seeding dev/staging with realistic, privacy-safe data instead of copying prod. - Load-testing with high volume, or crafting edge cases on purpose. - Do NOT use for production data generation or ML training data augmentation. ## Workflow ``` - [ ] Seed the generator for deterministic, reproducible output - [ ] Preserve referential integrity (child keys reference generated parents) - [ ] Include edge cases: nulls, duplicates, boundaries, late/out-of-order events - [ ] Match production distributions where behavior depends on them - [ ] Scale volume for load tests; keep small fixtures for unit tests ``` 1. **Deterministic + seeded.** Fix the random seed so tests are reproducible; flaky data makes flaky tests. Small, fixed fixtures for unit tests. 2. **Referential integrity.** Generate parents first, then children referencing real parent keys — otherwise join/relationship tests are meaningless. 3. **Edge cases on purpose.** Include nulls, duplicate keys, boundary values, empty batches, and late/out-of-order timestamps so pipelines are tested against what actually breaks them. 4. **Realistic distributions** where logic depends on them (skew, seasonality) — uniform random data hides skew bugs. 5. **Privacy-safe.** Synthetic stand-ins let you test without copying PII (pairs with `masking-pii-data`). ## Patterns **Deterministic, referentially-cons