Control clocks, randomness, providers, storage, and concurrency for repeatable tests; inject deterministic dependencies at the boundary.
Sources of Nondeterminism
- Time, time zones, calendars, date cutoffs, settlement windows
- Random ids, UUIDs, references, sampling, shuffling
- External providers: payments, market data, bank files, email, SMS, KYC
- Background jobs, queues, scheduled tasks, retries, polling
- Database ordering without explicit
ORDER BY - Parallel writes, locks, optimistic concurrency, race-prone state
Test Controls
Prefer existing patterns, else smallest boundary: frozen-time clock, deterministic id factory, provider fake (success/decline/timeout/duplicate), in-memory queue/sync runner, fixture reset, stable sort. Freeze time, seed ids, drive providers from fakes, assert observable behavior.
Prove Stability
Run repeatedly with the repo's runner:
npm test -- <target>
pytest <target> -q
go test ./... -count=20
Guardrails
- Do not add sleeps to fix timing, or hit real payment/market-data providers from tests.
- Do not assert on unordered collections without sorting, use wall-clock dates in expected values, or over-mock internal collaborators.