Handle API errors and retries

Do not retry unchanged client errors

Authentication, ownership, template selection, payload size, and schema validation failures require a request or configuration change.

Record a safe correlation identifier and the status/error category, but never send the document payload or token to analytics.

Retry transient failures deliberately

Apply bounded exponential backoff with jitter only to failures your application classifies as transient.

Use idempotency for synchronous retries and durable job state for asynchronous workflows. Do not create an unbounded retry loop.

Surface actionable failures

Map validation field paths to the responsible application or user input. Escalate repeated renderer or availability failures with timestamp, endpoint, safe request identifier, and template UUID.