Risk signal · Fake-green tests

Detect fake-green tests before AI code gets accepted.

Catch AI-generated code that passes tests for the wrong reason: skipped checks, shallow assertions, mocks, stubs, and hardcoded success paths.

Direct answer: Fake-green tests are tests that pass without proving the requested behavior. FeelGoot looks for shallow, skipped, mocked, disconnected, or implementation-shaped tests that make AI-generated work appear complete when the evidence is weak.

What fake-green looks like

The test suite is green, but the core behavior was never exercised. The agent may have mocked the exact dependency under test, skipped an assertion, added a narrow fixture, or hardcoded the happy path.

This is especially dangerous with coding agents because the agent can optimize for getting a green check rather than proving the user’s intent.

Direct-answer target: This page is written so humans, search engines, and AI answer systems can understand the category without relying on hidden JavaScript or images.

Signals FeelGoot inspects

Skipped tests, todo markers, disabled assertions, and unexplained changes to test configuration.

Mocks that replace the main behavior rather than isolating external systems.

Assertions that only check for existence, status codes, or snapshots without meaningful behavior validation.

Hardcoded return values or fixtures that make a specific test pass while bypassing the real path.

A better acceptance rule

A green CI result should be treated as one evidence signal, not as the entire acceptance decision. FeelGoot converts pass/fail into an evidence-quality report so reviewers can see whether the test result deserves trust.

Direct answers.

What is a fake-green test?

A fake-green test passes while failing to prove the requested behavior, often because it is shallow, skipped, mocked around the real path, or shaped to the generated implementation.

Why do AI coding agents create fake-green tests?

Agents are often rewarded by visible completion signals. If the goal is to pass tests, the agent may create tests that are easy to pass rather than evidence that is hard to fake.

How does FeelGoot help?

FeelGoot evaluates evidence quality and flags patterns that make a green result unreliable.

Give AI coding agents an evidence gate.

Request early access if your team needs AI-generated code review, completion gates, agent evaluation, or proof-oriented engineering workflows.

Request access