---
url: https://game-engine-io.github.io/enginex-frontend-sdk/concepts/validation.md
---
# Response validation

The SDK validates every gateway response with a schema and converts `snake_case` to `camelCase`.
With `validation`, you choose what happens when the response deviates from what is expected:

| Mode | Behavior |
|---|---|
| `'warn'` (default) | Calls `onValidationIssue` and delivers the converted response; only the diverging fields arrive as they came. |
| `'strict'` | Throws an `EnginexError` with code `VALIDATION_ERROR`. |

```ts
createEnginexClient({
  validation: import.meta.env.DEV ? 'strict' : 'warn',
  onValidationIssue: (issue) => reportToSentry('enginex-contract', issue),
})
```

## What `'warn'` mode delivers

Recovery is **per field**: a field outside the format arrives as it came (with keys in
`camelCase`), and the whole rest of the response goes through the normal conversions — unwrapped lists
(`pixKeys.list()` still returns an array), normalized enums (`'pending'`, not
`'WITHDRAWAL_STATUS_PENDING'`), money as `number`, protobuf defaults filled in, empty `nextCursor`
omitted. A required field that did not come is left absent. `onValidationIssue` receives the full paths
of the diverging fields (e.g. `['keys', 0, 'verified']`).

If the response's **root** does not have the expected format (e.g. text or a list in place of an object, or
none of the possible shapes of a response that has more than one), there is nothing to recover: `'warn'`
mode also throws `VALIDATION_ERROR` (with the problems in `error.details`), and
`onValidationIssue` is not called.

The `issue.raw` delivered to `onValidationIssue` is a copy of the body with the values of sensitive
keys replaced by `'[REDACTED]'`, at any level and case-insensitively:

* **credentials**: any key that contains `token`, `password`, `secret` or `backup_codes`;
* **personal data**: any key that contains `cpf`, `email`, `phone`, `name` or
  `birth_date`/`birthDate`;
* **device and network**: keys with the exact name `ip`, `device`, `user_agent` or `userAgent` (e.g.
  the sessions from `profile.listSessions`); `device_id` or `description` do not match;
* **URLs**: keys ending in `_url` or `Url` (e.g. `avatar_url`, `gameUrl`), which may carry
  tokens in the query string.

This way you can send the whole issue to Sentry without leaking tokens or passwords. The data delivered to the app
is not altered.

::: tip LGPD
Sending a response body to a third-party service (Sentry, Datadog, a centralized log) is
processing of personal data. The masking above covers the most common personal fields in the gateway's
responses, but it matches by **key name**: personal data under a key with another name (e.g. an
address) goes through as it came. If your log destination cannot receive any personal data, send only
`issue.route` and the `path`/`code` of `issue.issues` — without `issue.raw`.
:::

## Default warning (without `onValidationIssue`)

Without `onValidationIssue`, `'warn'` mode (the default) still warns: once per route per client, via
`console.warn`, with the route and the paths of the fields that did not match — **never** the response
values. Configure `onValidationIssue` to replace this warning (e.g. send it to Sentry instead of
the console).

Responses that carry tokens or secrets (`login`, session refresh and `enable2fa`) are always
validated in `'strict'` mode, whatever the client's `validation` is: a response outside the
schema becomes `VALIDATION_ERROR` and nothing is written to the session.

New fields the gateway starts sending are discarded until the SDK declares them.
