Crowdin measurement source
Collect Crowdin translation activity in Radar and use defined measurements to examine the text and editing process.
Keep producing and reviewing translations in Crowdin while using Lingo.dev to examine their quality. The Crowdin connection receives supported string-level events and makes their text available as measurement evidence in Radar. Your translations and publishing workflow stay in Crowdin.
In the Studio 2.5 navigation, open Studio 2.5 → Integrations instead of Settings → Integrations. Existing connection and setup links still work.
A source change, a new suggestion, an approval, and an update to exported target text describe different parts of the work. The connection retains those distinctions so a measurement can examine the text or change it actually needs. It does not generate translations or send corrected wording back to Crowdin.
Connect the source #
- In the intended Lingo.dev organization, open Settings → Integrations and select Connect for Crowdin. If it is unavailable, check the organization's feature access.
- Enter a Name for the connection. Leave Enterprise domain empty for crowdin.com. For Crowdin Enterprise, enter the part before
.crowdin.com, such asacme, then select Connect. - Copy the supplied webhook URL. In each Crowdin project you want to measure, open Tools → Webhooks and add a webhook using that URL, the POST method, and JSON payloads. Select the supported string events below.
Treat the webhook URL as a credential: the receiver uses its secret to authorize incoming events. Keep it out of tickets and repository files. One organization connection can receive events from several projects; each event carries its project identity. Record which projects you configured so you know what a later report covers.
Select the string events you need #
Subscribe to supported string-level events in Crowdin:
| Crowdin events | Evidence supplied |
|---|---|
string.added, string.updated, string.deleted | Source-string activity. |
suggestion.added, suggestion.updated, suggestion.deleted | Candidate-translation activity. |
suggestion.approved, suggestion.disapproved | Review actions on a suggestion. |
translation.updated | The change between old and new exported target text. |
For an exported translation, an empty old value and a present new value indicate creation. Two present values indicate an update; a present old value and an empty new value indicate deletion. Approval of a suggestion is a separate event from a change to exported target text.
File-level events such as file.approved and file.translated do not become string evidence through this adapter. Unsupported event types and events missing project or string identity can be skipped. Use a supported string event for the first verification rather than treating any successful webhook delivery as proof of usable evidence.
Follow one translation change into a measurement #
Use the built-in edit distance measurement for this first test. It runs on target creations and updates and compares the new target with the string's previous translation. Enable the measurement in Radar before creating the test activity.
- Select a known project, string, and target language that already has a translation. Record that translation and the text you will approve.
- Approve an edited suggestion through Crowdin's normal review workflow, and inspect delivery of
translation.updatedto the configured webhook. - Find the corresponding target event in Lingo.dev. Compare its project, readable key, locale, and target with the recorded pair. If the previous translation the measurement used is not the intended comparison, check the string's earlier events before interpreting a distance.
- Open the edit-distance result. Confirm the definition, compared texts, and completed numeric answer. The built-in measurement reference explains its denominator and character-unit limits.
This test connects one provider action to the answer you will later interpret. The first translation of a string has no earlier version, so edit distance answers it with 0. The approval itself is recorded as suggestion.approved and triggers none of the edit measures. The built-in glossary and rule checks consume completed-localization events, not Crowdin edits or approvals. To measure another event type, configure a custom measurement for that exact event and its available evidence; built-in triggers cannot be reassigned.
Event timestamps and identities help associate a record with its source, but their meanings depend on the activity. Translation and suggestion events, including approvals, use the translation's createdAt; an approval received today can appear under an older creation date. Receipt time distinguishes repeated disapprovals of one suggestion and substitutes for missing or invalid timestamps; an approval is recorded once per suggestion. Check the recorded eventAt when filtering Events or selecting a historical measurement period. Later text extraction cannot reliably rematch every timestamp-free event. Missing authorship also does not justify assigning an event to a human or model without evidence.
Read results with their coverage #
Webhooks collect configured activity from the time it is delivered. They do not establish every earlier translation, source revision, or project history. Missing payload text, unsupported events, and unconfigured projects can all reduce what is available to measure. Repeated review deliveries can produce separate evidence records, so event counts are not necessarily counts of unique reviewed translations.
Separate records that do not fit a criterion from applicable records missing required text. Among evaluations, distinguish completed answers, completed indeterminate answers, and work with no completed answer. An indeterminate answer records completed evaluation work but supports neither a pass nor a fail.
When reporting a pass rate, use determinate answers as its denominator and report how much eligible content received a determinate answer. A missing evaluation does not establish poor translation quality, and a Crowdin approval describes that project's review process rather than independently proving semantic correctness. Design a translation-quality measurement develops the calculation and interpretation.
Repair delivery or retire the source #
If evidence stops appearing, check the project's webhook subscriptions, event type, delivery errors, and destination URL. Confirm that it sends POST requests with JSON to the connection's current webhook URL. The connection does not automatically backfill missed events. To evaluate events already collected in Lingo.dev, use the historical measurement workflow.
If the webhook URL is exposed or needs replacing, open the connection in Settings → Integrations, choose Rotate, and confirm with Rotate URL. The old URL stops working immediately. Copy the new URL into every configured Crowdin project, then verify delivery of a supported event and its evidence in Radar.
Disconnecting stops new evidence from the connection; recorded measurements remain. Remove the Crowdin project webhooks separately. Crowdin translations and review actions are unchanged. For reports spanning a disconnect, account for the period without collection rather than interpreting it as a period without translation work.