A SECOND LOOK AT THE SMALL DETAILS
Two columns.
Read the reason.
A number warning asks you to compare two texts. A month name may become a number, a tag may contain an ID, or a quantity may actually have changed.
Start with a small table
Segment Review is a free browser beta for a separate literal check of two columns. It does not correct translations or approve a delivery. Keep your original table and review each reason in context.
Paste tab-separated text (TSV), with at least two columns. This is not an XLSX reader. Actual Excel clipboard interoperability and all CAT exports have not been validated. Quoted tabs, line breaks and doubled quote characters are supported; rows must have the same number of columns. Limits are 2,000 data rows, 64 columns and 1 MiB of UTF-8 text.
Use Try a sample to explore. Before starting your own table, choose Clear workspace, then paste it. For the original three-row example below, leave optional usage measurement off and select Source and Target; Reference is only for your own matching.
Reference Source Target
A Updated in March 2024. 2024年3月更新。
B Send 12 boxes. 21箱送る。
C Hello {name} Bonjour {nom}- Paste into Paste tab-separated bilingual text. Keep First row contains column names checked for this example, then select Read table.
- Choose Source column and Target column. Check the first three data rows in the preview. A misplaced column can produce plausible-looking but useless results.
- Keep Numbers as written and Named placeholders on. Leave Required phrases empty for this example. Select Review selected columns.
- Compare each reason with the original text and project requirements. Make any correction in your working translation separately.
Three reasons, three different decisions
The example produces three rows for review and three reasons:
| Table row | Reason | What to decide |
|---|---|---|
| A / 2 | The source contains 2024; the target contains 2024 and 3. | The standalone date is preserved. A literal difference alone is not an error. |
| B / 3 | The quantity changes from 12 to 21. | This is a deliberately wrong quantity in the example. Check against the source. |
| C / 4 | {name} changes to {nom}. | A translated token may break a placeholder contract. Verify the actual required token and format. |
Turning Numbers as written off leaves only row C for review. It hides both the legitimate date difference and the deliberately wrong quantity. Use the switch to choose the scope; it does not accept numeric translations for you.
No configured differences means only that the selected literal checks found none. Finish the linguistic, formatting and delivery checks required by your project.
Read the limits with the result
- Numbers: exact ASCII digit tokens and their counts, with optional signs and dot/comma separators. The rule does not interpret dates, units, number words or locale conventions. A valid
1,234.50 → 1.234,50change is listed. Digits inside markup are also text. - Known misses:
5 kg → 5 lbcan have no finding although the unit changed..7 → 7can also have no finding because the token pattern starts at a digit. Check these conditions separately. - Named placeholders: simple
{name}tokens, including repeats and letter case. The rule does not parse ICU messages, HTML/CAT tags or printf placeholders. - Required phrases: literal, case-sensitive substring pairs, one source phrase and required target phrase per tab-separated row, without a header; up to 100 pairs. No word-boundary or meaning inference. Empty targets are always listed.
If parsing fails, inspect the column count and quoting rather than deleting text just to make it fit. A line break inside a quoted cell stays in one table record; table row numbers are not physical line numbers.
Save a record and clear the workspace
Download JSON report saves all reviewed data rows, the selected source/target text, rules and reasons. The result list displays at most 100 rows in the selected view. Other input columns, including Reference, are not in the JSON; retain the original table to match its row positions.
The report is not a CAT import, an approval record or a way to mark an exception as accepted. Changing text, columns or rules invalidates the old result; rerun before downloading again.
Clear workspace removes the app’s working copy. Previously downloaded reports remain on your device. The tool does not save your text in browser storage or upload it. This does not describe clipboard, browser or OS retention, or make the hosting provider’s delivery/security processing disappear.
Check the QA tool you already use
If you use Xbench, check your installed version against its official change log ↗. Its official QA instructions ↗ cover selected checks, Project/Personal Checklists, marking findings, result export and Ignore Tags. Those existing options may already fit your task.
Historical support records illustrate context, not unresolved bugs in today’s build:
- Month names, 2017 ↗: a month-specific checklist was suggested and the requester reported success. Confirm the required date convention and your current checklist.
- Tag numbers, 2017 ↗: Ignore Tags was suggested, with later reported success. Inspect original CAT tags separately; hiding a number does not repair a missing tag.
- Abbreviation and dot, 2020 ↗: the requester reported that upgrading resolved the issue that day. Update and retest before treating an old report as an active limitation.
We have not compared the current Xbench build in use here, established a lower false-positive rate, or measured time saved by this beta.
Your text stays out of measurement
The tool’s optional usage measurement starts off and is separate for that page. Only while checked, it sends a page opt-in, successful reads of your table and successful result displays, with a broad traffic category. These counts do not confirm that findings are correct, a translation is approved or a report was saved.
Source/target text, required phrases, error text, filenames, text lengths and full URLs are not sent. The built-in sample is excluded from table/result counts. To start your own table after the sample, choose Clear workspace first. Text you paste yourself follows the normal opt-in rules.
Consent and a random measurement token stay in memory for this page; there are no measurement cookies or saved consent. Records older than 60 days are removed when a new event is received. Uncheck to stop future measurement; previously sent counts remain. Consent and tokens are not carried to other tool pages. Hosting providers may process information for delivery and security. This guide itself has no usage-measurement script.