Styles that never compile
Tailwind classes assembled from template literals at runtime. The compiler never sees the string, so the style silently does not exist, and the hover state you wrote has never once fired.
hover:border-${color}Website QC reads the code and the rendered page, finds the specific defects these generators leave behind, and explains why each one matters. It is closer to a linter with visual awareness than to a design opinion.
Six identical blocks, because there were six things to say. Nothing on the page is more important than anything else.
repetition · no focal point
Each finding names a specific element and a checkable cause, not a general impression.
One thing leads, the rest support it, and the page ends where it should. Same content, different weight.
These are defect classes traced back to how the generators work, which is why they recur across unrelated projects rather than appearing once.
Tailwind classes assembled from template literals at runtime. The compiler never sees the string, so the style silently does not exist, and the hover state you wrote has never once fired.
hover:border-${color}Logos and photos served from an unrelated or abandoned repository rather than your own project. One deletion by someone else and your live site has a broken image where the logo was.
A primary and a secondary action labelled differently but wired to the same destination, so a new visitor clicking "get started" lands on a sign-in form asking for credentials they do not have yet.
No footer, or a section the rest of the codebase already assumes exists, left out by an incomplete generation pass and never noticed because the page still renders.
A leftover component from an earlier attempt still resolving ahead of the one you actually edited, which is why your change appeared to do nothing.
An external image with no error handling, degrading to the browser's default broken icon in front of a customer instead of something you chose.
Send a repository or a live URL and it gets reviewed properly. Two validation passes are complete against real public repositories: one on Lovable-family output, one on a Bolt.new project, both finding checkable defects and producing patches that built and type-checked clean with no regressions.
The intended shape is a system that watches a repository and proposes small, specific patches a maintainer would actually merge. That is the direction, not the current state, and nothing on this page should be read as saying it exists today.
No aggregate metrics are published here because none would be honest yet. There is no "sites analysed" number, and there will not be one until it means something.
Tell me what generated it and what you already suspect is wrong. The review is manual right now.