| A worker dies mid-run | trigger.dev retries the run; the SDK finds each finished step’s checkpoint and skips it. | trigger.dev and the SDK | A worker crashes |
|---|
| An outside API answers 503 | Backoff of 1 s, 2 s, 4 s, then the dead-letter queue and a person. | The SDK’s retry policy | An outside API fails |
|---|
| A job is delivered twice | Its idempotency key is already claimed, so the side effects don’t happen again. | The SDK and run state | A job arrives twice |
|---|
| One city floods the desk | Its own queue runs 5 at a time; the other city’s jobs don’t wait behind it. | trigger.dev concurrency keys | One city sends 50 jobs |
|---|
| A model provider is down | The gateway falls back to gpt-oss-120b and reports the switch and its cost. | LiteLLM | A model provider is down |
|---|
| An agent loops | The run’s key runs out and the gateway refuses the call; the engine’s 12-call limit is the second stop. | LiteLLM and the SDK | An agent loops |
|---|
| A document gives the model orders | It’s quarantined: shown in the results, kept out of the prompt and never cited. | The index and the recommend step | A document gives orders |
|---|
| A person says no, or never answers | The waitpoint ends the run as rejected, or as expired after 5 minutes. | A trigger.dev waitpoint | A person says no |
|---|
| The model answers in the wrong shape | One more try; then the step fails as a validation error. | The SDK | Unit tests |
|---|
| The model approves over an unmet rule | Code refers it instead and names the rule. | The workflow | Unit tests |
|---|
| The gateway is asleep | The run’s first call waits up to about 60 s for the container, and the wait is traced. | Cloudflare Containers | Unit tests |
|---|