Korean product UX benchmark

Source

The source pair is synthetic. The initial source.html fixture belongs to the shipped korean-fintech review pack. The after.html fixture corrects only the three static form defects the pack can confirm.

Change

The phone field gains type=tel and autocomplete=tel, the password field declares current-password, and optional marketing consent is no longer preselected. The revision does not claim that payment disclosure, dense financial layouts, Korean wrapping, or assistive-technology behavior passed.

Verification

Run:

design-ai benchmark korean-product-ux --strict

The benchmark requires the initial report to contain the three namespaced confirmed findings and the revised report to contain none. It also requires payment disclosure, density, runtime accessibility, and the general runtime check to remain unverified. Any missing or unexpected finding is a regression.

Permission boundary

The Korean review pack is selected explicitly. Locale alone never activates it. The command reads packaged HTML and pack definitions only; it performs no write, browser run, target-repository mutation, or external request.

Remaining risk

This case proves deterministic form semantics and honest uncertainty. It does not prove legal compliance, payment completion, responsive Korean copy, density under real data, or accessibility in a rendered product. Those require scenario and browser evidence owned by the product team.

Claim boundary

This synthetic fixture does not prove customer adoption, Korean legal compliance, or production usability. It proves only the review pack's declared static rules and the continued visibility of evidence gaps.