Every string, every language, and what state it is in.
Who is waiting on what, which rows failed which check, and what the reviewer changed. It is all in the workspace, written by the run itself, so there is no spreadsheet to keep up to date beside the product.
Progress lives in a spreadsheet that is out of date the moment you send it.
What is ready, by language and by key, in the workspace, with the four checks recorded on every row.
Terminology drifts, because the glossary is a document nobody opens.
The glossary is enforced when the engine writes and again at review, with up to 40 terms folded into each chunk.
Review happens without context, so length and layout break after sign-off.
In-context review shows the rendered screen, so length and layout are judged against it rather than guessed at.
Four numbers, and where each one comes from.
deterministic checks on every row: placeholders, plurals, glossary and length. Source: the QA runner pre-gate.
MQM dimensions in the AI review pass: accuracy, fluency, terminology, style and locale convention. Source: the MQM typology.
CLDR plural languages, so a language that declares six forms is held to six. Source: the generated plural category table.
days of in-context screenshot retention, on 15 minute review sessions. Source: the in-context configuration.
FAQs
What a localization manager and a reviewer both ask on day one.