Limits and known gaps

What XRA does not do, where captures degrade, and how to tell before you ship.

A cloner that only advertises its wins wastes your time. These are the real edges.

Live applications do not come across

XRA reproduces a page, not the software behind it. A live demo, a logged-in dashboard, a search that queries a backend, a chart driven by an API — all capture as the markup that was on screen at that moment. If a section of the origin is an application, expect a still.

JavaScript-driven layout degrades

Refine strips the origin's JavaScript so the output is editable. Anything whose position depended on that JavaScript can land wrong: elements animated into place, layouts measured at runtime, dual-row headers that assemble themselves. Content is present in the DOM but may not sit where it did.

Interaction replay is not state

Recorded interactions replay a captured DOM change. That covers menus, dropdowns, modals and hover reveals. It does not cover anything with real state — carousels that advance, live search, steppers, forms that validate.

Route discovery is link-based

Pages are found by following links. A site that routes entirely in JavaScript, with no crawlable links and no sitemap, yields its entry page only unless you seed routes explicitly.

Icon-only controls need a label

A hamburger or search button with no text, no id and no stable class can only be re-found by its accessible label. Where an origin ships none, replay for that control will not bind.

Fidelity is measured, so check the measurement

Every run counts failed requests across every route. Read that number. A build can score well overall and still be missing an entire webfont family — which is exactly how a clone ships looking subtly wrong.

Licensing is your responsibility

The tool does not know whether you have the right to a design. Cloning a site you do not own can infringe copyright and trademark, and shipping one publicly with its branding intact is the clearest way to find that out.