Radar

Follow translation activity from the text collected to the measurements, explanations, and recurring problems it reveals.

When a translation receives a low score, the team needs more than the number. Which words caused the finding? What requirement did they miss? Was the evaluator looking at the version the team actually uses?

Radar connects collected translation activity with defined measurements. Open an event to see the available source, translation, and context, then follow its measurement answers. This makes it possible to examine a suspected problem before deciding whether to correct the text, clarify the requirement, or investigate the evaluator.

For example, an approved translation in Crowdin can become an event in Radar. A custom measurement can ask whether it preserves the source's conditions. The answer describes the collected text under that criterion. An approval in Crowdin and a passing measurement answer record two different decisions; examining both helps explain what the team accepted and what the check found.

Follow collected translations #

Start with Collect translations to connect supported activity or choose local scoring for files. Translation production can remain in the existing workflow. Configure a measurement before generating the activity that should trigger it, then inspect one event and its answer.

TaskReference
Find the text and activity received from a sourceLogs
Look up what an event type, producer, or answer meansReference
Turn checks on for selected events and languagesMeasurements
Understand edit distance, terminology, or rule checksBuilt-in measurements
Express a requirement specific to the contentCustom measurements
Review translations your localization engines produceReview Lingo.dev translations
Apply a current check to already collected eventsHistorical evaluation
Interpret an answer, N/A, or missing resultResults

Radar has five pages: Overview, Explore, Logs, Measurements, and Reference. Logs, measurements, and the reference are available to organization members; the overview's report tiles and Explore require the reports entitlement. A measurement selects event types and narrows them by language, source language, localization engine, and sample, whether the events come from a connected source or from Lingo.dev itself.

Investigate translations generated in Lingo.dev #

For a generated translation, the request record explains which model and guidance were actually used. Current settings alone cannot establish that history: a glossary or model choice may have changed since the request ran.

Use Runs to locate a submission and its locale-level work, then Requests to inspect attempts, pipeline stages, output, and errors. These investigation screens are currently under Studio. Recovery stays with the API, CLI, or integration that submitted the work.

Measurements review those translations too. Each completed localization is a localization.completed event, so a measurement that runs on it examines the output. Review Lingo.dev translations covers the engine's Automatic review switches and conditions limited to particular engines. The answers appear with the event in Radar → Logs and in the engine's Logs tab.

Investigate a pattern #

Reports cover the overview and saved Explore reports over events and measurement answers. A report can show where to look more closely; individual findings explain what happened there. Keep the count of examined translations beside their outcomes so missing evidence remains visible.

Measure introduces the broader method: define what must remain true, examine answers, and establish how much content the evidence describes.