Skip to main content

​
The Problem

AI-generated tests look correct and pass — because they mock the thing they are testing. Coverage metrics lie. Production defects escape through fully-mocked, 100%-green test suites.

​
What It Does

Three-stage quality backstop:
  1. Audit — detects test smells: mocking the unit under test, asserting only on mock args, no real execution path
  2. Generate — produces test plans from acceptance criteria (real execution, no fake-green)
  3. Validate — runs symbol-level smoke tests pre-deploy

​
Try It

Audit an existing test suite: 
activate nablr
Set agent to tester. Audit the test suite in tests/ for smells and fake-green coverage
Generate tests from AC: 
Set agent to tester. Write tests for STORY-001 based on the acceptance criteria
Pre-deploy smoke check: 
Set agent to tester. Run symbol-level smoke tests before deploy

​
Expected Output

  • Test smell report — per-file violations with severity
  • AC-derived test plan — real execution paths, no mocked units
  • Smoke test results — pass/fail per critical symbol

​
Who This Is For

  • QA leads inheriting untested or mis-tested codebases
  • Teams with green CI and recurring production bugs
  • Engineering managers introducing AI coding assistants and needing a quality backstop
TestArchitect rejects tests that mock the unit under test. A test that only asserts on mock.call_args is not a test — it is a false signal.