Every one of the 212 documented operations has a retry, idempotency, and completion policy. Policies are generated conservatively from the OpenAPI contract. Endpoint-specific business error codes are listed only where the operation description documents them; they are never invented. Every operation also uses the common error contract.
For state-changing operations, an HTTP timeout is an unknown outcome—not proof that the operation failed. Query current state before retrying.
The machine-readable source is the Operation Behavior Manifest.
| Product | Operations | With endpoint-specific error codes | Matrix |
|---|---|---|---|
| Browser | 53 | 19 | Browser |
| Cloud Phone | 109 | 50 | Cloud Phone |
| Cloud Storage | 24 | 12 | Cloud Storage |
| Shared Resources | 26 | 12 | Shared Resources |
| Policy | Meaning |
|---|---|
| Safe with bounded exponential backoff and jitter. | Read-only operation. Repeating it does not change resource state. |
| Query current state before retrying after a timeout or transport failure. | The operation may already have been applied. Read the resource before sending it again. |
| Do not retry automatically; first query whether the resource, upload, or task was created. | The operation creates something and has no idempotency key, so a blind retry can duplicate it. |
| Do not retry automatically unless endpoint-specific documentation says it is safe. | The contract does not say enough to classify the operation. Treat it as unsafe. |
Generated coverage: 212 operations; 93 have endpoint-specific documented business codes and 119 currently rely on the common error contract.