Changelog

One line per change, newest first. The same list is under your name in the app, What's new, where a workspace sees only the entries for the features it has.

November 2026

5 Nov Analytics has a new section, Quality over time: the weekly score, findings by severity and the share of findings people confirmed or ignored, by profile, language and account. Free shows 4 weeks, Solo 12, Team and above 52 with a CSV download. PMs and admins see the workspace; a profile's owner sees that profile.

5 Nov A profile's owner can now make rules from words and add terms to that profile whatever their role. A translator who owns a profile now reaches Profiles. Only PMs, admins and the profile's owner can change a profile. The new rule editor is on in every workspace.

5 Nov In the report, Show tag contents (key t) shows each inline tag as it is in the file instead of a numbered chip. It changes only what you see; the file and its copies stay as they are.

5 Nov Inconsistencies are shown as a group. The report row reads the shared text and "3 translations · 14 segments"; the pane lists each translation with its segments and marks the most used one; a segment number opens that segment; Make all like this writes one translation into every segment of the group through the report's editor and keeps any segment whose tags differ. Excel and CSV list each translation with its count and segment numbers.

5 Nov Edge and double spaces: warning in software string projects (an XLIFF whose file is a .strings, .resx, .json or similar table) and short UI segments, all languages, info in prose; a native string file gets the same warning for double spaces. Insert now restores or removes the edge space; a trailing newline is tolerated only in Phrase exports.

4 Nov Chrome extension 0.7.38: Insert now works in Phrase, including editors built on Phrase, as it does in WorldServer. In XTM Cloud and Smartling it is still in testing and Copy works as before. In Smartling, a check reads every page of a long job; if it cannot, the card says which strings were checked. A job you can only view offers Copy only.

4 Nov Chrome extension 0.7.37: in Phrase, Auto-detect before a check now names the profile the project's name points to, as the check itself does. A fix you undo can be inserted again. The card keeps its place, is taller and can be read by screen readers.

4 Nov Profiles that never saved their 100%, ICE and Locked boxes now skip ICE matches and locked segments, as the profile screen has always shown. 100% matches are still checked. The report says how many segments were left out and why. To check them, untick ICE or Locked on the profile. A changed segment the file marks do not translate is always reported.

4 Nov QA API: skipExactMatches: false and skipLocked: false now check those segments even where the profile skips them; before, a false was read as not sent. The result of /check and /check-segments has a new skipped field with the number of segments left out for each reason.

3 Nov Chrome extension: in Phrase, Copy now copies a fix as plain text without tag markers, ready to paste into the open segment.

3 Nov Pricing: Solo is £19 a month with 50,000 AI-reviewed words, Business is £499 and includes LQA scorecards on three client templates; nobody already subscribed pays more.

October 2026

31 Oct Narrated how-to videos now sit at the top of 12 Help pages. Each one shows the steps for a feature, from your first check to an LQA scorecard, with captions and a transcript.

30 Oct In a Dropbox folder run, segments outside the role's scope, such as ICE and locked segments for an editor, are now left out and not sent to the AI review, except a finding inconsistent with another segment.

30 Oct Uploaded files can now be kept longer for a workspace that agrees it with us: we set the number of days per workspace, the corrected copy can be downloaded for that long while its report remains available and the Security and data tab in Settings shows the number of days. Unset, uploads are deleted within 24 hours as before.

30 Oct QA API: a check sent through the API now runs the same rule checks as a check in the app. A fuzzy gettext .po entry whose translation belongs to an earlier source text is reported once as not in use, and with the AI review off the answer carries the two meaning checks that run without it, for the target languages they cover. A profile's list of checks now names those two when the AI review is off, for the same languages.

30 Oct Turkish: the negation check reports a Turkish finding only when the AI confirms it, "must not" no longer reads as "must", the consistency check reads curly and straight apostrophes as one and an English target may pluralise a brand.

30 Oct Retention in Settings: a workspace admin now sets how many days QA reports and finalized LQA scorecards are kept in Settings under Security and data. The card says what the next nightly sweep will delete before it runs and a PM sees it read only.

30 Oct Chrome extension 0.7.35: Run starts a check only from your own click, an upload that stalls is tried once more when the extension is connected to your own account, without starting the check twice and a slow editor page no longer leaves the extension waiting.

30 Oct QA by email: when a brief DNS failure stops us checking who sent a message, the message is rejected with a failure notice instead of being dropped without a word. Resend it after the DNS issue clears and it runs.

30 Oct The whole-file term check no longer reports a product name kept in English, or a word that happens to stand next to the term, as a second translation. It names the one rendering the file uses most and the one that differs.

30 Oct A source translated two ways is now a major finding; a repeated target whose sources differ only by a number or a placeholder is no longer reported, as the number and placeholder checks own it.

30 Oct Japanese and Chinese: a space between two katakana words and a sentence that ends in 。 where the English ends in a colon are no longer reported.

30 Oct gettext .po files: a fuzzy entry whose translation belongs to an earlier source text is reported once as not in use and is not checked against the new text, and plural forms are read with the file's own plural rule, so an untranslated plural is one finding and a zero form is not a number error.

30 Oct A corrected target that only rewrites a number in the target's own format is now offered as a fix.

30 Oct Two meaning checks now run without the AI review: a prohibition that lost its negation and a quantity or obligation word turned into its opposite are reported by the rule check and marked as confirmed when the AI review agrees.

29 Oct The security pack: Security and privacy in Help is now a section with seven pages, one per question a security review asks: the controls we have and do not have, where your text goes when a model is used, the sub-processors with their regions, backups and recovery, support and incidents, status and uptime and the data processing agreement. The public security page carries the facts a procurement reviewer needs without an account.

29 Oct Software string files with bare CR line ends: each finding now names the right line, and a .po file keeps its multi-line values and line ends in the corrected copy. An assistant no longer offers a corrected copy for a zip whose only string file has nothing to pair with.

29 Oct Retention per workspace: an admin sets through the API how many days QA reports and finalized LQA scorecards are kept, and a nightly sweep removes the older ones with their files and decisions. Unset keeps everything as before.

29 Oct The report marks the one place a fix is written in the target, not every place the same word stands. The extension's card marks the words a fix changes.

29 Oct Corrected copies for software string files: a .po file, or a zip with the source and the translated .json, .resx, .strings, .stringsdict or .properties file, comes back with only the changed values written; everything else stays byte for byte.

29 Oct JUnit and SARIF output for CI: a run's findings download as JUnit XML and SARIF 2.1.0, so a pipeline step fails on the findings you choose and annotates the file by key. The Help has recipes for GitHub Actions and GitLab CI.

29 Oct The audit trail now also records who exported a report, downloaded or sent a corrected copy, decided on a finding, exported LQA scorecards as CSV or changed the dispute policy or the template rules. A deleted workspace's trail is deleted with it. The Help names every action, under Developers, QA checks, runs and reports.

29 Oct TER beside chrF++ in Eval and the edit rate in Compare's CSV and Excel: two numbers for pricing post-editing against a reference.

29 Oct Every workspace now keeps an audit trail of who signed in, changed a role, made or revoked an API key or a webhook, signed off a report and changed an LQA finding, dispute or scorecard. Each row holds the action and the fields that changed but never your text. The rows form a chain you can verify. A workspace admin reads the trail and downloads it as CSV through the API.

29 Oct Software string files, part two: .NET .resx, iOS and macOS .strings and .stringsdict and Java .properties files are checked like any bilingual file, named by key. Give each with its source file, together or in a zip.

28 Oct Software string files: gettext .po and i18n .json files are checked like any bilingual file, named by key. Give a .json file with its source file, together or in a zip.

28 Oct LQA disputes and MQM core: a workspace can choose who resolves a dispute, keep the person who raised it from resolving it and give each one a deadline, and the API lists the disputes that are open, overdue or resolved. A dispute raised by a signed-in reviewer now names that reviewer. Two new templates use the MQM core categories and a workspace can pick a template by content type.

28 Oct The usage answer of the API and of an assistant now lists the words each AI pass sent this month. Only the AI review counts against your allowance, as before.

28 Oct A server with no AI provider set up now says so: every check runs the rule checks only and the report reads "AI checks are not set up on this server. Only the rule checks ran." instead of an AI failure.

28 Oct Bulk CSV in the API: your LQA scorecards, findings and disputes and your Eval runs as CSV for Excel or your BI tool. The analytics answer is never cut short; a query too wide to answer says so.

28 Oct 100% and ICE are now two separate skip options on Run a check, a profile, LQA and Compare, so you can skip one and still check the other.

28 Oct No-break spaces are kept: the double-space fix no longer removes a no-break space, a missing no-break space before French punctuation now has a fix you can accept and the report shows which space in the evidence is a no-break one.

28 Oct Learn reads QA reports by content: an Xbench or other QA report in any column order is recognised, its words and format marks become Learn cards and a QA report dropped on Run a check is sent to Learn instead.

28 Oct Turkish number formats: an English amount or thousands-grouped number copied unchanged into a Turkish target is reported with the Turkish form, and an apostrophe after a name that starts with İ is checked correctly.

28 Oct The whole-file review no longer repeats a segment's own text as a finding and keeps to the segments it was given.

28 Oct LQA analytics by date and group in the API: your finalized scorecards for any date range, by month, language, account, content type, project, vendor or translator, with errors per 1,000 words, the pass rate and the disputes.

27 Oct Fixes on tagged segments: a fix may now change the words between several tags and move a tag with its words, in the report, the corrected file and an assistant. A tag is never dropped, repeated or added.

27 Oct Send the corrected copy back: from the report's Download menu, save it beside the original in Dropbox or email it to the person who sent the file. A Smartling run sends your findings back as issues and the API gives the copy in one call. It is always a new file and your original is never changed.

27 Oct Source checks: when the AI review reads a word that is not in the source but looks like one that is, the finding now asks you to check the source, as a minor finding with no fix to accept. A German heading translated as a noun phrase or as an instruction is no longer reported as an omission.

27 Oct Tracked changes for Trados, a new option: download the corrected SDLXLIFF with each fix as a revision you accept or reject in Trados Studio. Check the file in Trados Studio before you send it to a client.

27 Oct Corrected copies of packages: the report and an assistant now write your fixes into the files inside a WorldServer, Trados, memoQ or zip package. Every other file in the package stays exactly as it is. Your original package is never changed.

27 Oct Corrected copies of Excel and CSV files: the report and an assistant now write your fixes into an .xlsx, .xlsm, .csv or .tsv file, changing only the target cells. Formatting, formulas and the rest of the file stay as they are. Your original file is never changed.

27 Oct Insert a fix from the Chrome extension: press Insert on a finding and the fix goes into the segment you have open in WorldServer, as if you typed it. You still save it in the editor and the report marks the segment Inserted. In the other editors you copy the fix.

23 Oct Edit in the report: on the Solo plan and up, change a segment's target in place or accept a suggested fix, then download a corrected copy of your file. Your original file is never changed.

23 Oct Whole-file review: on the Solo plan and up, the AI can read each file as a whole for terms, labels, names, headings and lists written two ways, one finding per problem with the segments to fix. It starts once each check is measured.

23 Oct File-wide checks: a term or UI label translated two ways, mixed heading styles, step numbers, formats and the form of address across a file are now one finding each with the segments to fix, on by default under Checks.

23 Oct A brand written in another form than your list, Iphone for iPhone or Pay-Pal for PayPal, is now its own finding, Brand written differently, with the list's form and a fix that keeps the suffix, so a re-run of an older job can show casing slips it passed before.

22 Oct Settings, AI review lists the providers first and the models follow your ticks, a minor finding can carry a weight in the score, Workspace shows the fallback profile in use and Analytics counts AI words, never money.

22 Oct A rule finding the AI agrees with or disputes now reads AI check of rule findings; Second opinion means only a second model rechecking the AI's own findings.

22 Oct The API documentation is now one guide in Help with every endpoint and limit, and Help reads well in dark mode.

22 Oct Home opens on your clean rate over the last 30 days with where the findings come from and the work that needs you, each chart with one line on what it means.

22 Oct Cmd+K or Ctrl+K opens Search or do on any keyboard layout, even while you type in a field.

21 Oct Build from a folder takes the folder itself: choose it or drop it and your browser packs it. A zip still works, and every kind of zip is picked on a Mac.

20 Oct Build a profile from a folder: drop a client's files and approve what it found.

17 Oct Dropbox folders. Connect a Dropbox account in Settings, Connections, watch a folder and every bilingual file dropped there is checked with the report written next to it. Team and up.

15 Oct Confirmed holds what the AI confirmed and what an exact rule check found.

15 Oct Deterministic checks no longer get a second opinion and a disputed finding counts at half weight until you decide.

15 Oct Profiles are easier to read: seven sections, each opening with its state, the rare settings under Advanced.

15 Oct The AI section asks one question, Standard or Premium. It shows what each costs in AI words. The model list is gone from profiles.

15 Oct Guidance lines are a readable list you can edit in place.

15 Oct Learn opens with what is waiting and one Add sources button.

15 Oct Checklists and custom rules are one list. Import a checklist, write a rule or describe one in your own words, in any language; they all run the same way and export together.

14 Oct The AI reads every translated segment now, including the ones a rule already flagged. The report says how many it read.

14 Oct Custom rules are one sentence each. Write when the source has X the target must not contain Y, test it on a file, paste a list or copy an Xbench checklist into rules you can edit.

14 Oct Spelling runs for 67 languages. Where no dictionary exists yet the AI alone looks for misspellings and the report says so.

14 Oct Each finding can show the segment before and after it, the same context the AI saw.

14 Oct Consistency findings say which segments they disagree with.

14 Oct Send a file to your profile's email address and the report comes back in the same thread.

14 Oct Claude and ChatGPT can run checks in your workspace once an admin connects them under Settings.

14 Oct AI guidance in a profile is now one list: each line shows where it came from (typed, a style guide, Learn, a report or the LQA brief) and can be switched off. Free text goes in the prompt addendum.

14 Oct Compare now sits under What's Wrong in the menu, after Analytics.

14 Oct The buttons that teach a profile from a report now read Keep as brand and Fine as is.

14 Oct An offer code from an event now gives Team free for a period when you sign up, with no card.

14 Oct The New menu opens with the n key and the letter beside each job starts it; no shortcut uses Cmd or Ctrl any more, so none can close your browser tab.

14 Oct Your account menu now opens from your name at the foot of the left menu, and only there.

14 Oct Spelling findings now carry a severity set by the AI. They are listed under Spelling and never count toward the result or the score.

14 Oct Spelling no longer flags a suffix attached to a placeholder or a tag as a misspelling, such as the Turkish ending after a tag.

14 Oct "Also on this segment" now shows the whole message of the other finding instead of cutting it short.

14 Oct Home now opens on the drop area, your five latest reports and what the AI caught this week; the menu groups every page under the product it belongs to.

14 Oct Reports lists each report as one row with its result, its score and when it ran; you can filter it to Critical open or Failed.

14 Oct The report learns from your decisions. Ignore a finding in a file and the next run of that file starts with it ignored; the teach row under a finding sends a suggestion to Learn; findings you ignore often appear there on their own.

13 Oct The report now says plainly what the AI did: it reads every translated segment and gives its opinion on each critical or major rule finding, and AI findings carry no second opinion.

Oct Your own systems can now run files, find and download reports, mark findings and share report links through the API, as the Reports page does.

Oct The word and finding counts now agree between the Reports list, the report and its download; a Dropbox run shows the folder its file came from.

Oct The AI words used this month stay in the top bar on every page, and Analytics and Integrations follow your plan.

Oct The second opinion now sees the tags, numbers and links a rule found before it agrees or disputes a finding.

Oct A changed or missing link or email address is now reported once as what it is, no longer also as a missing placeholder.

Oct After an upload, Run a check shows the file's format, segments, words, language pair and locked segments before you run it.

Oct A workspace admin can now delete a single report for good, after confirming it.