Test Data Management

Most flake and most false positives trace back to data: shared rows, leftover state, a clock that moved, an ID that collided.

Factories, not fixtures

A fixture file is a snapshot of assumptions that rot. A factory is a function with defaults and overrides — the test states only what it cares about:

const user = userFactory({ plan: 'pro' })              // everything else is a sane default
const order = orderFactory({ user, items: [item({ cents: 1000 })] })

Read the test and you see exactly which fields drive the assertion. That is the point.

Isolation ladder (pick the highest rung you can afford)

1. In-memory / pure — no shared state at all.
2. Transaction per test, rolled back after. Fast, total isolation, no cleanup code.
3. Namespace per test — unique tenant/org/prefix per test, parallel-safe.
4. Truncate between tests — slow, order-sensitive, last resort.
5. Shared seeded database — where flake is born. Migrate away.

Parallel workers each need their own namespace or their own schema. A worker id in the tenant name (tenant-w${process.env.TEST_WORKER_INDEX}) is usually enough.

Determinism

• Clock: inject it. vi.setSystemTime, freezegun, or a Clock port. Never new Date() inside code under test without a seam. Test the month-end and DST cases explicitly.
• Randomness: seed it, and assert on the seeded output.
• IDs: deterministic in tests (uuid('order', 1)), so failures are readable and diffs stable.
• Ordering: never assume database return order. Assert on sets, or order explicitly.

Sensitive data

Never copy production data into a test environment without anonymization — it is a breach waiting for a screenshot. Generate realistic-looking data instead (@faker-js/faker, factory_boy, gofakeit), seeded so it is reproducible.

Cleanup

Prefer strategies that cannot leak: rollback, ephemeral namespace, containerized database per run. If you must delete, do it in a fixture teardown that runs even when the test fails, and verify the teardown ran — a silent teardown failure looks exactly like a flaky test tomorrow.