Use cases / illustrative scenarios
Where drift changes the outcome.
These scenarios explain policy decisions. They are synthetic examples, never customer incidents or measured recovery results.
Synthetic scenario / 01
A request field is renamed.
A vendor expects payment_method where a tool still sends source. A selective request patch can change that field while preserving unrelated values.
- Prerequisites
- Authoritative field identity and type evidence, an explicit trusted repair engine, confidence and verification gates, and write policy. Nango codegen separately requires a caller-confirmed same-type sibling rename.
- Limitations
- The default guest has no selective repair. Shape passthrough may still be rejected by upstream. No automatic retries or write execution.
Synthetic scenario / 02
An amount changes units.
A cents-to-dollars shift can change meaning without changing the JSON type. A 200 response does not prove write integrity. Unrepairable semantic writes fail closed in repair and strict modes; observe only records.
- Prerequisites
- Explicit scoped statistical windows with sufficient baseline/live samples, compatible profiles and precision budgets, or a configured deterministic invariant.
- Limitations
- No universal conversion or detection guarantee. A response decision cannot undo an already completed write. Meaning changes without an observable signal can remain undetected.
Synthetic scenario / 03
A response loses a field.
A read result no longer contains the expected status. Reads always fail open: preserve the original outcome and report available drift evidence. Internal engine faults bypass inspection.
- Prerequisites
- An explicit detector and baseline, a supported complete capture and configured event listeners. Retained artifacts need opt-in audit capture, host drain and a durable sink.
- Limitations
- Unknown-length, chunked and non-JSON Fetch bodies skip inspection. Missing evidence is unavailable, not a clean result. No fabricated field values or production recovery claim.
Decide with the evidence.
Payload verification does not prove write integrity. Show write-integrity evidence separately. Review the original result, dispatch status and idempotency binding before deciding whether to replay a write.
Read the protection and adoption model →