KvasirQA 1.0.0 — manual

Free guide download? The KvasirQA-1.0.0-guide-and-examples.zip download contains this manual and the practice files only. The steps below need the app (KvasirQA.html and its app folder), which is in the paid download KvasirQA-1.0.0.zip.

KvasirQA reads bilingual translation files and lists problems a reviewer or client would notice: untranslated segments, numbers, tags, links and e-mail addresses that differ from the source, glossary terms that were not used, the same sentence translated in different ways, and punctuation and spacing slips. It never changes your files and never uploads them: it runs in your browser from a folder on your computer. You fix what it finds in your translation tool, export again and check again.

1. What you need

• Desktop Google Chrome. This release was tested with Chrome 152 on Linux (x86-64).
• Windows, macOS, other browsers and phones have not been tested for this release.
• Bilingual files, up to 50 MB each: XLIFF 1.2 (including segmented files), XLIFF 2.0 or XLIFF 2.1 (.xlf, .xliff), or TMX 1.4 (.tmx, the TMX 1.4b format). The version and namespace are checked; other versions are refused with a message.
• No account, licence key, internet connection or installation.

Tested with synthetic XLIFF 1.2, 2.0, 2.1 and TMX 1.4 files written to the published specifications. Real Trados, memoQ and Phrase exports have not been tested; vendor-specific files and packages may not work. Free alternatives such as Okapi CheckMate cover similar checks; KvasirQA offers a smaller browser workflow.

2. Open KvasirQA

1. Extract the whole download. Unzip KvasirQA-1.0.0.zip into a folder you can find again. Keep KvasirQA.html and the app, docs and examples folders together.
2. Open KvasirQA.html in Chrome. Drag the file into a Chrome window, or right-click it and choose Open with › Google Chrome.
3. Check what you see: the heading Check translations before you deliver them, a green-edged box Start here: a 2-minute practice run, and three numbered panels.

3. Practice run with the included files

The practice set describes a fictional kettle brand, “Brightfern”: a segmented English → German XLIFF 1.2 file, an English → French XLIFF 2.1 file, an English → German TMX 1.4 memory and a three-term glossary. The mistakes are planted on purpose.

1. Click Load practice files.
2. Expected result: 3 files, 29 segments checked, 7 errors, 6 warnings. The three files are listed in panel 1 and the glossary shows brightfern-glossary.csv · 3 terms.
3. Click Download CSV or Download HTML to save the report (see Save the report). You have finished the practice run.

The 13 planted findings:

Check	File › segment	What is wrong
Numbers	kettle manual (de) › 3	“1.7 litres” became “1,5 Liter”.
Tags	kettle manual (de) › 8	Placeholder tag x:2 is missing in the German text.
Links and e-mail	kettle manual (de) › 9	The support link was changed from /support to /hilfe.
Untranslated	kettle manual (de) › 10; app (fr) › hold/s2	No target text.
Glossary	kettle manual (de) › 6 and › 7	“power button” should be “Ein-/Aus-Taste”; “descaler” should be “Entkalker”.
Same source, different translations	de: manual › 4, › 6 and memory › m1; fr: app › start and › start-again	“Press the power button.” and “Start boiling” were translated two ways.
Same translation, different sources	app (fr) › filter, › filters	“Filtre” is used for both “Filter” and “Filters”.
Target same as source	kettle manual (de) › 12	“Brightfern” is unchanged — correct for a brand name. Warnings can be intentional.
Ending punctuation	kettle manual (de) › 11	The source ends with “.”, the target has no full stop.
Spaces	kettle manual (de) › 11	Double space in the German text.

4. Read the findings

• Tiles count files, segments, errors (likely mistakes) and warnings (review them; some are intentional).
• Chips filter by check. Click several to combine them; click All to reset.
• Search filters by words in the source, target, file name, segment or explanation.
• Each grey row names a file, its format and languages. Below it, Where gives the segment (and the unit, for segmented XLIFF), What to fix explains the problem.
• Consistency checks compare all loaded files with the same language pair (source and target codes, for example en-US → de-DE). In the example below, the German file disagrees with the translation memory.
• The line under the tiles states the scope: files and segments checked and units skipped. If a file could not be read, or units could not be checked (a TMX unit without a defined source language, or a unit with <sub> text inside an inline code), it turns amber and says Partial result; the files tile shows “N not checked” for unread files. Counts and reports then cover only what was checked, and every report lists what was left out. Units marked translate="no" are listed too, but do not make the result partial.

On a narrow window the table becomes one card per finding:

5. Save the report

Button	File	Use it for
Download CSV	kvasirqa-report-<date>-<time>.csv	Excel, LibreOffice or Google Sheets. UTF-8 with a byte-order mark; one row per finding. Cells that start with =, +, - or @ get a leading apostrophe so spreadsheets do not run them as formulas.
Download HTML	…html	A standalone page with files, counts per check and all findings. Open it in a browser and print or save as PDF for a client.
JSON	…json	Automation and archiving: every finding with file, segment, languages and details.

Chrome saves the files in your usual downloads folder. The report reflects the checks that were switched on when you clicked the button.

6. Check your own files

1. Click Clear all.
2. Export the bilingual files from your translation tool as XLIFF 1.2, XLIFF 2.0/2.1 or TMX 1.4 (for TMX, with a specific source language).
3. Drop them on Drop files here or choose files, or click choose files. You can add several files at once; a file with the same name replaces the earlier one.
4. Optional: Choose glossary… (see below) and switch checks on or off in panel 2. Ignore upper/lower case when comparing translations makes “Speichern” and “speichern” count as the same translation.
5. Fix the findings in your translation tool, export again, add the new files and check that the counts go down.

Units marked translate="no" are skipped and counted in the file card: on an XLIFF 1.2 trans-unit or an enclosing group, or on an XLIFF 2.0/2.1 unit, group or file. The nearest setting wins, so a unit marked translate="yes" inside such a group is checked. Segments locked in your tool by other means are checked like any other segment.

7. Glossary file

A CSV or TSV file (comma, semicolon or tab; UTF-8) with one term per row:

source,target,language
power button,Ein-/Aus-Taste,de
descaler,Entkalker|Entkalkungsmittel,de
descaling,détartrage,fr

• The header row is optional. language is optional: de applies to de, de-DE, de-AT…; empty means every target language.
• Separate allowed translations with |. Matching ignores upper/lower case and only matches whole words.

8. When something goes wrong

Problems are shown in red under the file or glossary concerned. Other files are still checked.

You see	Cause	Do this	It worked when
Buttons do nothing, page has no colours	Opened from inside the ZIP, or the app folder is missing	Extract the whole ZIP again and open KvasirQA.html from the extracted folder	The green Start here box appears
“not well-formed XML (line …, column …)”	The file is damaged or was edited by hand	Export it again from your tool, or fix the XML near that line	The red message disappears after you add the file again
“Not a bilingual file: the main element is <…>”	You added the original document or another XML file	Export the bilingual XLIFF/TMX instead	The file card shows its format and segment count
“No checkable segments were found”	The export is empty, a TMX has no defined source language, or every unit has <sub> text inside an inline code	Check the export settings in your tool; export a TMX with a specific source language	Segments are counted
“Nothing to check in this … file: … marked translate="no"”	Every unit is marked not-for-translation, directly or by an enclosing group or file	Nothing, if that is intended; otherwise export the units that need translation	The file card shows its segment count
“Unsupported XLIFF version” / “Unsupported TMX version”	The file is XLIFF 1.0/1.1/2.2 or TMX other than 1.4, or its namespace does not match its version	Export as XLIFF 1.2, 2.0, 2.1 or TMX 1.4	The file card shows its format
Amber “Partial result” line	Some files were not checked, or units were skipped that KvasirQA cannot check (TMX source language undefined, <sub> text inside an inline code)	Fix or re-export the files listed, then add them again; review skipped units in your tool	The line starts with “Scope:” instead
“not valid UTF-8” / encoding message	The file was saved in another encoding	Re-export as UTF-8	The file card shows its format
“larger than 50 MB”	Very large file (for example with an embedded original document)	Split the project or check fewer files at a time	The file is read
Glossary message in red	No rows with both a source and a target term	Use the source,target,language columns	“… · N terms” appears
“No findings match this filter”	Chips or search hide every finding	Click All or clear the search box	Rows appear again

Support: give the KvasirQA version, your operating system and browser, the step and the exact message, and whether the practice files work. Do not send client files; if needed, send a small made-up example that shows the problem.

9. Rules used by each check

Segment text. Text inside formatting and highlight markers (XLIFF 1.2 g, mrk; XLIFF 2 pc, mrk; TMX hi) is kept. Native code (bpt, ept, ph, it; TMX also ut) is left out of the text and counted as a tag. Units whose source or target has <sub> text inside such a code (an XLIFF 1.2 or TMX sub-flow, for example a footnote or a link title) are skipped, counted and the result is marked partial: KvasirQA does not check sub-flow text. translate="no" on an XLIFF 1.2 trans-unit or enclosing group, or on an XLIFF 2.0/2.1 unit, group or file, skips the unit (nearest setting wins). Inline translate="no" markers inside a segment (XLIFF 2 mrk/sm) are not used: that text is checked like the rest of the segment. Whitespace runs are collapsed to one space for comparisons. Segments: segmented XLIFF 1.2 uses seg-source/mrk mtype="seg" (segment = mid); if the target of such a unit has text but no segment markers, the whole unit is checked as one segment; otherwise one segment per trans-unit. XLIFF 2: each segment of a unit. TMX: the source language is the unit's srclang if present, otherwise the header's; every other language variant is a target. Units whose source language is *all*, missing, or not among their variants are skipped and counted, never guessed. Not checked: XLIFF 1.2 alt-trans alternatives and XLIFF 2 ignorable content.

Check	Severity	Reported when
Untranslated	error	The source has text and the target is missing or empty. Other checks skip such segments.
Numbers	error	The digit sequences differ. A number is digits with optional . , ' or non-breaking-space separators; separators are ignored, so 1,234.5 = 1.234,5 = 1 234,5 (with a non-breaking space). An ordinary space splits numbers; 3.5 and 3,50 differ.
Tags	error	The inline codes (type and id) in source and target differ in kind or count. Order is not compared. Compared: XLIFF 1.2 g, x, bx, ex, bpt, ept, ph, it; XLIFF 2.0/2.1 ph, pc, sc, ec; TMX bpt, ept, it, ph, ut. Highlight and annotation markers (TMX hi, XLIFF mrk, XLIFF 2 sm/em) are not compared: a target that drops a hi marker is not reported, but the text inside it is checked.
Links and e-mail	error	http(s):// addresses or e-mail addresses differ (trailing punctuation ignored).
Glossary	error	The source contains a glossary term (whole word, any case) and the target contains none of its translations.
Same source, different translations	warning	Identical source text has more than one translation among all loaded files with the same language pair (source and target codes; upper/lower case ignored; de and de-DE are different codes).
Same translation, different sources	warning	One translation is used for more than one source text (same language pair).
Target same as source	warning	The target equals the source and contains letters.
Ending punctuation	warning	After closing quotes and brackets, the last character differs among . ! ? : ; …, or only one side ends with one. Full-width 。！？：；, Arabic ؟ ؛, Urdu ۔ and Devanagari । count as their equivalents; for Greek (el) a final ; counts as a question mark.
Spaces	warning	The target has a double space that the source does not, or leading/trailing spaces differ.

10. Limits and privacy

• KvasirQA does not judge meaning, grammar, style or spelling, and a clean report does not prove a translation is right.
• It does not write back to your files or to your translation tool, and does not read CAT-tool statuses, comments or locks.
• Not supported: other XLIFF/TMX versions, PO, TS, TTX, DOCX/RTF bilingual tables and compressed project packages. Export XLIFF 1.2/2.0/2.1 or TMX 1.4 first.
• Sub-flow text (<sub> inside inline codes) is not checked: such units are skipped and counted, and the result is marked partial. Highlight markers (TMX hi, XLIFF mrk) are not compared by the Tags check; the text inside them is.
• Files over 50 MB are refused. The table shows the first 500 findings; the CSV, HTML and JSON reports contain all of them.
• Tested only in desktop Chrome 152 on Linux with synthetic files; other systems, browsers and real tool exports are untested.
• Privacy: files are read by your browser on your computer. KvasirQA makes no network requests (its page blocks them) and stores only one setting in the browser: whether the guide is hidden.

KvasirQA 1.0.0 · Licence: LICENSE.txt · Changes: CHANGELOG.md