Quality Checks That Travel With the Work
Every localization team has a list of things a job must get right: the glossary applied, the client style guide followed, the product names untouched. Until now that list tended to live in emails and project notes, retyped for every job. This release makes it part of the workflow. You can now build reusable job checklists in Settings, under Project and work processes, each with a name, an internal note suppliers never see, and a set of requirements marked optional or mandatory. Attach a checklist to a step in a workflow template or on a project's own workflow, and every job created on that step carries the same requirements. Edit the checklist once and every job using it picks up the change. Checklists work in both Standard and Codyt projects, and they slot naturally into the workflow automation you already run.
Automated checks got sharper too. Four new built-in QA rules catch technical tokens that change by accident during translation: camelCase terms such as getUserName, fully uppercase words such as MAX_RETRIES, web and email addresses, and alphanumeric codes such as AB123X4. Each rule compares source and target in both directions, so a token that was altered, dropped, or added gets a precise message during the regular QA pass. For software, gaming, and technical documentation, where one mangled variable can break a build, that is the difference between catching an error in review and hearing about it from a customer. The rules are configured per QA profile, including a minimum length for the uppercase check, and all four are off by default so existing profiles behave exactly as before. The QA categories reference lists every option, and our quality assurance overview shows how the checks fit together.
Machine Translation in Your House Style
Machine translation is most useful when its output already sounds like you. The DeepL connector now supports DeepL Style Rules: an administrator can load the style rules defined in the DeepL account and choose one per target language, right next to the existing Glossaries option. Translations into that language then follow your organization's writing preferences automatically. Style rules are managed in DeepL, require a DeepL Pro API subscription, and are currently offered for German, English, Spanish, French, Italian, Japanese, Korean, and Chinese. A connector with no rule selected works exactly as it did. Setup steps are in the DeepL connector documentation, and the wider picture is on our machine translation page.
Post-editors also get a clearer view of where a translation came from. On a machine-translated segment, the MT badge and the name of the engine that produced it are now shown to every user, including external translators and reviewers who previously saw neither. Knowing which engine wrote a translation helps a post-editor judge how far to trust it.
Image to Text, the OCR feature that pulls text out of images, has moved to Microsoft's supported vision service ahead of the retirement of the version it used before. Existing settings and keys carry over, so nothing needs reconfiguring, and language coverage grows substantially: printed text goes from 25 to 164 languages, and handwriting from English only to nine languages. Extracted text can come out slightly different, usually better, and images smaller than 50 pixels are no longer accepted because the new service will not read them.
AI workflows became more dependable in two places. When a job assigned to an AI agent starts on its own, for example because the previous job in the workflow finished, the AI work now starts reliably, and if a hand-off ever fails the job history records it instead of leaving the job sitting at In progress with nothing running. AI post-editing, translation, and quality runs also complete on files where a sentence contains a quoted phrase, with the quotation marks preserved.
Suppliers See What They Need, and Only That
Good supplier relationships depend on sharing the right information at the right moment. In Standard projects with multi-step workflows, a supplier reviewing a proposal for a dependent job, such as a review that follows translation, can now see the previous job's expected deadline before accepting. A note explains that the job depends on work not yet complete, and the "Previous job deadline" line gives the date. Suppliers can decide whether the timing works without the project manager sending it separately. This covers allocated proposals, group job openings, and crowd jobs, as described in the guide to accepting and declining Standard jobs.
Confidentiality got the same attention. In a Standard job opening offered to several suppliers at once, each invited supplier now sees only their own company's messages and the project manager's, never a competitor's, in the messages dialog, the "Last message" preview, and the job tooltips. Standard jobs now match the privacy Codyt jobs already had. When an in-house client cancels an order, the automatic job history entries and the cancellation emails sent to suppliers now come from the project manager rather than from the client's own name and address, so client contact details stay with the agency. And the new supplier group form now offers only job types that are currently enabled, so a group can no longer be tied to a job type that will never receive work. Teams managing large supplier networks will feel all of these in daily operations.
Fewer Accidents, Better Data Out
Some changes are about protecting you from a single wrong click. Setting a project to Completed, Archived, or Cancelled now asks for confirmation first, and each message names what that status can set in motion, such as client notifications, automatic archiving, or the cancellation of open jobs and quotes. Declining leaves the popup exactly as it was, comment text included, and every other status still applies on the first click. This applies to both Standard and Codyt projects.
Integration teams get better visibility. The Webhook Calls page under Settings, then Activity, now has filters for URL, outcome, and creation date, remembered between visits, so you can find a specific call without paging through the whole list. Any successful response code, such as 201 or 204, now counts as a success rather than a warning, and the page states that webhook calls are kept for seven days; the system activity documentation has the details. The Jobs list endpoint of the Wordbee API can now also return the edit-distance figures Wordbee records for finished jobs, showing how much the translator changed and how far the final text ended up from the raw machine translation. The four new columns use the same names and meanings as the matching report columns, so what you pull through the API matches what the report shows.
Also in This Release
One more fix works quietly in the background. Accounts with a very large number of price lists could overload a shared server cache and slow things down for everyone on that server. Pricing data is now stored more efficiently and the cache rebuilds far more quickly, so nothing changes on screen, and the first price-list load is faster for every account.
The full technical notes for 2026.3.10.0 live on help.wordbee.com, with configuration steps for each item. If job checklists or the new QA rules could change how your team runs quality, talk to our team and we will help you set them up.