> ## Documentation Index
> Fetch the complete documentation index at: https://help-center-starter.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Error handling

> Retries, compensating steps, and alerting when workflows fail.

Workflows fail in the real world: APIs time out, databases lock, humans go offline. Good error handling turns chaos into retries, compensations, and clear operator signals instead of silent data drift.

## Failures

Classify transient versus permanent failures. Transient steps should retry with jitter; permanent failures should stop fast and surface a crisp error code.

### Observability

Attach correlation IDs across steps so support can trace a single user action through every hop. Log HTTP status bodies at reduced verbosity to avoid leaking secrets.

### Poison messages

If the same payload fails repeatedly, stop retrying and quarantine it—otherwise you starve healthy traffic.

## Compensation

Undo partial effects when a downstream step fails—especially for payments or external posts. Compensation may mean voiding an invoice, deleting a draft record, or sending a corrective webhook.

### Ordering

Design compensations in reverse dependency order: undo the last successful side effect first.

### Partial success

When only one of two external systems succeeded, document the manual reconciliation path in the alert body.

## Alerts

Route failures to on-call channels with runbook links. Include workflow name, step name, and last payload hash—not full PII—in the first line of the alert.

### Fatigue

Throttle duplicate alerts for the same root cause. Pair alerts with dashboards so responders can see whether failures are trending or isolated.