Send Human Effort Only Where It Is Needed
The headline change in this release rethinks what a workflow step actually has to touch. In Standard projects, a step can now be set to work on only the segments whose status matches a set you choose, red-status segments for example, instead of the whole document. Wordbee waits until the previous step finishes, then builds the job from just those segments and counts the words at that moment. Until it is ready the job shows a short "not ready yet" message, and if nothing matches the chosen status, the job cancels itself rather than sitting empty.
That is a small setting with a big effect on cost and turnaround. A human reviewer opens a job that contains exactly the content that needs them and nothing else, so the expensive part of the process, expert attention, lands where it matters. You can shape workflows around the state of the content rather than the size of the file, and the counts stay honest because they are taken at the moment the work is really handed over. The setup steps are covered in the guide to dynamic content selection jobs.
Two supporting changes keep automation predictable. When you enable an AI-flow parameter that only works with a concrete value, such as a password, number, checkbox, or dropdown, Wordbee now asks you to fill that value in before it will save, so you can no longer switch an option on and quietly leave it doing nothing. Free-text fields like Custom Instructions stay optional, and existing configurations keep working untouched. And AI-powered machine translation no longer reports a failure just because a flow is taking longer than usual to finish. The run now waits for the flow's real outcome instead of giving up early on work that would have completed on its own.
Truer to the Source File, Whatever the Format
A localization platform earns trust one file at a time, and this release tightens how several formats come in and go back out. For XLIFF, a new option reads each file's declared target language when you mark files online and routes each one only to its matching project language, so a German file lands in the German document and a French file in the French document. You can tick every language at once without creating wrong-language duplicates. A second new XLIFF option keeps non-translatable rows visible instead of dropping them: rows that are pure markup or explicitly marked "do not translate" appear as locked, green, read-only rows, giving translators full context while keeping the technical content safe from edits and out of QA's way. Both options are off by default and work the same for XLIFF v1 and v2, and you turn them on in the XLIFF configuration options.
Offline round-trips are cleaner too. When you re-import an XLIFF file, untouched 100% pre-translations now come back neutral gray instead of flagged red, matching how the stronger in-context matches already behave. Valid reused content from your translation memory no longer arrives looking like an error that someone has to clear by hand, while the online QA check still catches anything genuinely wrong.
The same care extends to Excel and InDesign. Multilingual Excel deliverables now decide their left-to-right or right-to-left layout by looking at all of the file's column languages together, so a sheet mixing Arabic with English and French stays left-to-right and flows only the Arabic text right-to-left within its own cells, no longer depending on which delivery line you finalized from. Full details are in the guide to translating multilingual Excel files. The InDesign parser gains an "Extract hidden conditional text" toggle, off by default, so text a designer has hidden with conditional text is excluded from translation and linguists stop spending time on content that will never appear in the final file; the reconstructed file always preserves the conditional text wrappers, so nothing is lost, as described in the notes on Adobe InDesign files.
Two fixes round out the fidelity story. Excel cells whose only content is a placeholder or variable are no longer dropped from multilingual and by-column deliverables; the original content is always restored. And previewing or building the source file of a document no longer overwrites its real source version, which had been quietly breaking later actions such as "mark for no translation."
A Clearer View While You Work
Good tooling should tell you the truth at a glance, and a few changes make the editor and jobs pages easier to read. When per-job segment status is enabled, the Editor's Statistics figures and the delivery "Translations done" checklist now count segments by the status of the job you are working in, matching what the segment rows already show. This closes a gap that could warn a translator about errors belonging to a different job, and it pairs naturally with the per-job progress view introduced last month. For the people doing the work, the numbers finally match the screen in front of them.
The Jobs pages help right from load. The work-progress bar now fills in with each job's real progress the moment the page opens, including jobs worked in the previous editor, so completed jobs no longer show an empty bar until you refresh each row. It is accurate on load for both Standard and Codyt jobs. And the Codyt jobs page now reopens on the sub-tab you last used rather than always defaulting to one view, and shows a short "No jobs in this view" message instead of a silent empty list, so a linguist always knows where they are and why a list looks the way it does.
Better Numbers and Tighter Access for Managers
Managers get more to work with in this release, starting with reporting. The per-job and per-segment edit-distance figures, how much a translation was changed and how far it moved from the raw machine translation, in both characters and percentage, are now available in reports alongside the name of the machine translation engine used. Nothing is recalculated; the reports simply read values already stored on each job, so teams who need these numbers can add them to custom report templates. The measures are explained in the guide to job edit distance.
Administrators also get finer control over single sign-on. You can now exempt specific companies from having their profile details updated by SSO at login: users from an excluded company still sign in normally, but their name, email, role, and company details are left untouched, which is useful when those details are managed by a separate integration. Everyone else continues to have their profile refreshed from SSO as before. This sits alongside the rest of the platform's security and access controls, and the settings are documented under single sign-on.
Also in This Release
A few smaller refinements round out 2026.3.3.0. On the many pages that use a second-level link bar, the active sub-tab is now underlined so you can instantly see which section or scope you are viewing. And three business reports, the C1028 Jobs Single Line Report, the S411 Oversight Report, and the S721 File QA Oversight Report, now generate reliably even when the selected date range covers a busy period with a large number of jobs, invoices, or files; their content and layout are unchanged.
The full technical notes for 2026.3.3.0 live on help.wordbee.com, with configuration steps and screenshots for each item. If any of these changes will reshape how your team works in Wordbee, talk to our team before you roll them out. We would rather walk you through setup than let a release stumble in production.