Evidence review
What to inspect before changing anything
Start with the delivered website and the real user journey. Record the current state so the team can distinguish an observed problem from an assumption and compare the same surface after deployment.
- 01Separate raising and lowering drivers
- 02Read raw values beside reference baselines
- 03Identify binary markers versus continuous structure metrics
Implementation
A practical improvement plan
Make the smallest coherent change that solves the observed problem. Keep normal code review, accessibility, security and product checks in the loop instead of optimizing for the scan alone.
- 01Use concrete drivers to frame a review question
- 02Change patterns only when product quality benefits
- 03Avoid optimizing directly against hidden model behavior
Verification
How to verify the result
Verification should test the intended outcome and the most likely regression. Use the production delivery path whenever headers, caching, rendering or third-party services affect the result.
- 01Rescan after one coherent change set
- 02Check whether the expected driver moved
- 03Review downstream design and performance effects
Common pitfall
A shortcut to avoid
Adding driver values together or subtracting them from the score produces a number the model does not define.
Further reading
Primary guidance and references
These sources provide standards, security guidance or the interpretation framework used to keep this guide bounded. Product-specific implementation still requires review in the actual codebase.