LocFit
Website localization QA, scored with MQM
Add the key pages of a site, get one ranked view of where its localization is breaking.

Most localization isn't meant to be perfect. It's meant to be good enough for the job. LocFit grades each page with the MQM framework, ranks them worst-first, and separates what's broken from what's just polish.

Pages to review
Each page is reviewed separately with the same MQM engine; the overview ranks the site’s pages worst-first. To save a live page as a file: open it, Ctrl/Cmd+S → "Webpage, HTML only".
Running this outside Claude? Add an API key

Leave blank when running inside Claude. Paste your own Anthropic API key to run LocFit on any web host.

Annotates errors per page, then computes each MQM score in your browser
Reviewing…
Something went wrong
Site localization health

Showing report for —

—

—/ 100 MQM quality scoreiScore = 100 − the per-word penalty. MQM severity weights: minor ×1, major ×5, critical ×25. One critical error is an automatic fail. On short samples the raw score swings hard, so penalty per 1000 words is the comparable figure.
—
—absolute penalty
—penalty / 1000 words
—words evaluated
MQM-Core scorecard — penalty by dimension
Error dimensionErrorsPenalty points
Total (APT)00
Must fix 0

Approving a fix here is a demo of the human-in-the-loop step. Write-back to a TMS / TMX is the planned next stage, nothing is synced yet.

Could polish 0