Verify a write after its response is lost
Design a read-back that can identify the intended operation even when the write response never reaches the executor.
A verification URL built from the write response cannot be constructed when that response is lost. If recovery matters, choose an identifier known before dispatch and a read endpoint that can locate the resulting operation by that identifier.
For declared HTTP actions, Anlyon can repeat a configured read-back during reconciliation when the rule does not need the missing response. A matching rule can confirm the outcome. A rule that depends on an unavailable response, or no verification rule at all, leaves the operation requiring operator evidence.
Compare two API designs
| Design | Recovery implication |
|---|---|
| POST creates a random ID returned only in the response | Lost response also loses the lookup key |
| Request includes a stable application operation ID | Read-back may locate the intended operation independently |
| GET returns only the current resource value | A match may not identify which write produced it |
| GET returns operation ID, target and result | Stronger evidence that the intended operation landed |
These are architecture examples, not assumptions about every provider's API. Verify what your chosen provider actually stores and returns.
Select a discriminating match
Use evidence created by the write. Matching a generic health response or a pre-existing status does not establish that this action ran. Include the operation identifier and consequential fields where the provider exposes them. Consider whether another actor can produce the same matched state.
Anlyon's declared verify rule performs a GET on the action's own host and checks configured JSON fields. It does not invent a provider lookup endpoint or automatically add provider idempotency. Keep those limitations visible in your action definition review.
Test the missing-response path
In a controlled fixture, make the write take effect and then drop the response. Confirm the read-back can still be constructed from known input. Also test no match and stale state: neither should become a fabricated success or an automatic new write.
The lost-response example demonstrates the reasoning locally. Use declared action setup and verification-rule design for the hosted configuration. Record provider-specific recovery limits before enabling production traffic.