---
title: "Verify a release check detects its known broken control"
description: "Evaluate a release regression test against externally prepared correct and broken sandbox builds, ensuring its verdict depends on the required outcome."
canonical_url: "https://vostride.com/agent-library/release-regression/known-failure-control-verdict.md"
md_url: "https://vostride.com/agent-library/release-regression/known-failure-control-verdict.md"
content_kind: "recipe"
topic: "release-regression"
---

# Verify a release check detects its known broken control

Evaluate a release regression test against externally prepared correct and broken sandbox builds, ensuring its verdict depends on the required outcome.

## Risk

A regression test that passes every build offers no evidence of detection. Introduce a bounded known defect in an isolated fixture and require the same unchanged assertions to distinguish it from the correct behavior.

## Fixtures

- Two owned sandbox build fixtures: a correct task-save flow and an intentionally broken variant that drops the task description while still showing success.
- One unchanged agent-qa test definition asserting reopened title and complete description, with a canonical test ID and configured targets.
- Independent task oracles and run artifacts for both fixtures; the broken build is prepared externally and never deployed to a real user environment.

## Steps

1. Run the unchanged test against the correct fixture and verify its completed verdict and reopened-record evidence support both saved fields.
2. Run the same assertions against the prepared broken fixture and verify the run reports failure on the missing persisted description.
3. Inspect the independent task oracle for each build and verify the correct record retains its body while the broken control demonstrably lacks it.
4. Compare run evidence and verify the failing verdict is tied to the intended persistence defect rather than missing credentials, unavailable targets, or a different setup failure.

## Expected results

- The correct fixture passes with observable persistence evidence.
- The broken fixture fails specifically because the required description was lost.
- Assertions and expected behavior remain unchanged between the two runs.

## False-pass traps

- An environment failure on the broken fixture does not prove the test detected the application regression.
- Weakening the description assertion until both builds pass defeats the detection control.

## Cleanup

- Delete both disposable task records and retire the intentionally broken sandbox build fixture.

## Use with agent-qa

Adapt this test design to your application's acceptance criteria and configured target using the [recipe execution guide](/agent-library/software-factory/turn-a-recipe-into-an-agent-qa-test.md). External fixture setup and non-UI checks require your own supported hooks or test tooling. Validate the authored test, execute it, and inspect the completed evidence; this recipe is not a recorded test result.

## References

- [agent-qa first test](/docs/agent-qa/guides/first-test.md)
- [agent-qa hooks](/docs/agent-qa/guides/hooks.md)
- [agent-qa test](/docs/agent-qa/configuration/test.md)

- [Release regression QA recipes](/agent-library/release-regression.md)
- [Software factory QA library](/agent-library.md)