Radar
A translation's story does not end when it is produced. A reviewer approves it, a person edits it in your repository or Help Center, a check finds a term it missed. Radar is where Lingo.dev collects that activity and examines it: every translation event is recorded, measurements answer questions about it - did it follow the glossary, how much did a person change it, does it meet your own criteria - and reports show where the answers form a pattern.
Radar covers the translations your localization engines produce and the translation work happening in tools you already use. A low score comes with the text it describes, so you can see which words caused it before you decide whether to fix the translation, clarify a rule, or question the check.
The four pages#
| Page | What it shows |
|---|---|
| Overview | This month at a glance: the average score of your measurements, words generated, events today, and the measurements that are on |
| Explore | Charts of events and measurement answers by language, engine, model, repository, and more, saved as reports for the organization |
| Events | Every event received, and every answer a measurement gave about one |
| Measurements | The built-in and custom checks, the events each runs on, and whether it is on |
Where events come from#
An event records something that happened to content: a localization completed, a commit changed a locale file, a suggestion was approved. Each carries the text it concerns - source, target, and the previous version where there is one - plus its locale and who or what produced it.
| Source | What becomes an event | How to connect |
|---|---|---|
| Lingo.dev localization engines | Each localization request and document job, as it starts and completes, with the source and target text, both locales, the engine, and the request's context | Nothing to set up |
| GitHub | Each localization key a commit adds, changes, or removes, in your source and target files | Connect the repository with the GitHub App. Its .lingo/config.json tells Radar which files are source and which are target |
| Zendesk Help Center | Each article published in the Help Center, as a source or translation change in each of its languages | Connect the Help Center |
| Crowdin | String-level activity: source changes, suggestions added, edited, approved, or disapproved, and updates to translated text | In Settings → Integrations, connect Crowdin with a personal access token, then add the webhook URL it gives you to each Crowdin project |
Connecting a source does not move your translation work into Lingo.dev. Translations keep being produced and published where they are today; Radar receives the activity and examines it.
Events#
Radar → Events has two tabs. Organic lists every event received; Measurements lists every answer a measurement gave. Narrow either to the last 7, 30, or 90 days, by language, and by event type; on Organic, By selects who produced the text - a model, a person, machine translation, translation memory, or an algorithm.
Open an event to see its type, locale, producer, actor, when it happened and when Radar received it, the content unit it concerns, every measurement answer about it under Values, and the stored Payload with its text and context.
The event type matters when you count. A suggestion, its approval, and the updated translation can all concern one string; counting them as three translations overstates the work.
Measurements#
A measurement is a check that runs when selected events arrive. Lingo.dev ships built-in measurements - glossary adherence, rule adherence, MQM, edit rate, edit distance, and edit patterns - and you can write custom ones in plain language, answering pass or fail or a score from 0 to 100. Each measurement chooses the event types it runs on and narrows them by language, source language, engine, and sample.
To review your engines' translations, switch on Glossary adherence and Rule adherence under Automatic review in the engine's Reviewers tab, or run a measurement on completed localizations. To see how much people change what Lingo.dev wrote, turn on edit rate and edit distance for a connected repository or Help Center. AI Reviewers covers every measurement, its conditions, and how to write a custom one.
Measurements run asynchronously. Their answers never block, change, or publish a translation.
Reports#
The overview follows the current month; Explore charts any question over events and answers and saves it as a report the whole organization sees. Reports explains each tile, how to build a report, and how its numbers are calculated.
Try it on one translation#
Turn on a check
Open a localization engine's Reviewers tab and switch on Glossary adherence under Automatic review.
Translate something
Find the event
Open Radar → Events, find the completed localization, and open it. The source and target text are in its payload.
Read the answer
Under Values, open the glossary adherence answer. Read its score and reasoning beside the text: which terms applied, and which the translation missed.
Access and plans#
Events, measurements, and their answers are available to every member of the organization, on every plan. Explore and the overview's report tiles come with the Enterprise plan. Where your plan meters review, each measurement answer about an event that meets the measurement's conditions is one billed run - see pricing. An event outside a measurement's conditions is answered N/A and is not billed.
Connecting a source follows that integration's plan availability.
Radar replaces Observability
AI reviews that lived under Observability → Measurements are now measurements in Radar → Measurements, and the earlier reports moved to the Radar overview and Explore. Individual translation requests and runs are still inspected in Studio.
Radar via MCP#
With the Lingo.dev MCP server, your AI assistant can work with Radar from the conversation: read events and their answers, create and tune measurements, query events and answers by language, engine, or repository, and save reports.
"Which languages had the lowest glossary adherence last month,
and show me three translations that failed it."