Type Inference Mental Model

DX rules of thumb:

• Token in → narrowed type out
  • Pass a literal entity token ('user' as const) to get token-narrowed record types at call sites (no generics).
• Domain vs storage
  • Reads return storage records (keys included). Use entityManager.removeKeys(entityToken, records) in handlers when you want domain shapes.
• Projection K channel
  • When you pass projection tuples (const), item shapes narrow at the call site. The QueryBuilder preserves this through the adapter.
• Values-first config narrows index/page-key types
  • Build your EntityManager via createEntityManager(config as const). The EntityClient carries the literal type, so createQueryBuilder infers index tokens (ITS) and per-index page keys automatically. Without a values-first literal, ITS defaults to string.

Patterns

// Token-aware get with projection tuple (const narrows the shape)
const out = await client.getItem('user', { hashKey2: 'h', rangeKey: 'r' }, [
  'a',
] as const);

// QueryBuilder narrows per-index page keys automatically when the client
// was constructed from createEntityManager(config as const)
const qb = createQueryBuilder({
  entityClient: client,
  entityToken: 'user',
  hashKeyToken: 'hashKey2',
});
Copy

Related

• Querying with QueryBuilder
• EntityClient: CRUD and batches