Existing product refactor benchmark¶
Source¶
The synthetic
before.html
fixture represents an existing settings page with four deliberate static defects.
after.html
changes only the markup needed to address those findings.
Change¶
The revision declares the document language and mobile viewport, gives the button a visible name, and gives the profile image an alternative. It keeps the page structure intact. This makes the case about evidence-led refactoring rather than a replacement redesign.
Verification¶
Run:
The runner validates both canonical quality reports. It compares the exact expected
and observed confirmed finding IDs before and after the change. Missing and
unexpected findings are listed separately; they are not converted into a quality
score. Runtime evidence must remain unverified in both reports.
Permission boundary¶
The runner reads the two packaged fixtures in place and never writes to them. It does not open a browser, execute the page, mutate a target repository, or contact an external service.
Remaining risk¶
The corrected source has no supported static defect, but keyboard order, screen reader output, focus appearance, loading, errors, repeated actions, motion, and rendered layout were not exercised. A browser evidence run is still required before calling the product behavior verified.
Claim boundary¶
This synthetic comparison does not prove customer adoption or production impact. It proves only that the deterministic inspector removes the expected static finding IDs without promoting missing runtime evidence.