Validation at the Boundary

Critical rules

• Validate untrusted input once at the edge (HTTP, CLI, queue, env, third-party). Everything inside trusts types by contract.
• Parse into richer types (Zod + branded types) — don't just boolean-check.
• Business functions contain no shape/format checks on args.
• Schema failures → VALIDATION_FAILED / HTTP 400. Business-rule failures → Results (result-types).
• Always parse third-party responses before use.
• Before schemas or branding choices, read references/examples.md and references/patterns.md.

Workflow

1. Identify every untrusted entry point for the feature.
2. Define Zod schemas; brand IDs/tokens that could be swapped by mistake.
3. Parse at the boundary (safeParse / middleware). Reject with a standard error envelope.
4. Pass inferred types into fn(args, deps) with no re-validation.
5. Keep domain rules (permissions, balances) inside business functions as Results.
6. For coercion, PATCH, transforms, and middleware, read references/patterns.md.

Resources

• references/examples.md — parse mindset, branded types, handlers, error format. Read when implementing boundaries.
• references/patterns.md — two layers, coercion, PATCH, middleware, rationalizations. Read for common patterns.

Validation

• Every external input parsed with Zod at the boundary
• Business functions accept validated types; no shape checks inside
• Confusable IDs/tokens use branded types
• Third-party responses parsed before use
• Invalid input → consistent VALIDATION_FAILED / 400
• Business-rule failures returned as Results
• No re-validation of internal or DB-sourced data

Constraints

• Do not validate between internal functions that already share a type contract.
• Related: fn-args-deps, result-types, api-design, strict-typescript, config-management.