Skip to main content

What the rules find

The rule pass compares the source and the translation for everything that has a right answer. It runs on every check and on every channel, and it never costs AI words. Each finding names the segment, shows the evidence (the actual tag, number or term that differs) and suggests a fix. Here is what it looks for, by category.

Structure and tags​

Whether the file parses at all and whether the segments line up. Inline tags and placeholders: a bold pair or a link dropped, added, changed or left unclosed, a variable such as {name} that the software fills in later lost or altered. A tag moved for word order is not flagged. This is the most common critical finding, because a lost placeholder breaks the page or shows the wrong value. When only plain formatting such as italic or bold is lost, the finding is "Formatting tags lost or added", lower than a lost placeholder, and a profile can change its severity. Font, style and theme names that a file keeps inside its tags are not read as words of the translation.

Numbers, units, currencies and dates​

A number that changed or went missing, a lost minus or plus sign, a currency symbol or code replaced by another, a measurement unit substituted, a date or number written in a format the target locale does not use, an alphanumeric code such as a part number not carried over.

Terminology and brands​

A glossary term translated the wrong way. A brand, product or programme name from Brands and terms to keep that was translated or left out. A brand that is there but written in another form than the list is Brand written differently: capitals changed (Iphone for iPhone), a Turkish dotted or dotless I (İntune), a name split or joined (Pay-Pal, FireTV). The finding shows the list's form against the target's and a fix that keeps the suffix. A segment left identical to the source that should have been translated, with the brands and the Fine as is list taken into account so a product name on its own is not flagged. The profile's own custom rules and checklists run here too.

A URL or an email address that changed or disappeared, unless the profile says that client localises them, in which case the right local address is checked instead.

Spacing and punctuation​

A leading or trailing space lost (so the next string is glued on), doubled spaces, a stray space before a closing mark, a mark typed twice, an unbalanced bracket or quote, a quote style the client does not use, punctuation at the end of the segment that differs from the source where it matters (a full stop or a colon the source does not have counts too). Two words glued together with a capital in the middle ("Güvenilirİş" for "Güvenilir İş"). Quotation marks of another locale, such as Swiss or French guillemets « » left in a Turkish target, and an apostrophe right after a closing quotation mark. A repeated word.

The edge-space checks (a space added at an edge, or the source's lost) run on XLIFF-family files. They and the double-space check are Major in a software string project (an XLIFF whose file is a .strings, .resx, .properties, .json or .po table, as Smartling, Phrase and Crowdin export them) and in short UI segments (six source words or fewer with no final punctuation; for Japanese, Chinese and Thai a lost edge space stays Minor), and Minor in prose, the same for every language; a profile that sets the rule keeps its own value. A native string file checked on its own gets the same Major for double spaces. Insert removes the extra edge space or restores the source's exact one. A lone line break at the end of a segment is tolerated only in a Phrase export, which writes one (a space lost before that line break is still reported and restored); anywhere else it is reported, without a fix.

Straight quotes are a client preference: a client that asks for straight quotes and apostrophes instead of “ ” ‘ ’ gets them checked when its profile switches on "Typographic quotes where the client asks for straight ones", or when its Xbench checklist with that check is imported.

Length and consistency​

A translation far longer or shorter than its source, a character limit exceeded. The same source translated two different ways in one file. A consistency finding says which segments it disagrees with.

Language specific checks​

Checks that only make sense for one language: a Turkish suffix that does not agree with the acronym it follows, the Turkish dotted and dotless i, French spacing before punctuation, Arabic punctuation and digit systems, right to left text that lost its script, mixed scripts in a Latin target, German register, emoji and flag sequences and more. Turkish has the deepest set. Each runs where the target language calls for it; the profile can turn any of them on or off.

The French spacing check offers a fix: it writes the missing no-break space inside « » and before ; : ? and the exclamation mark (Canadian French: before the colon and inside « » only).

Invisible problems​

Control characters and invisible marks in the target, markers a translation tool leaves behind, a translate="no" segment that was altered.

What you control​

Every check has a severity (critical, major, minor) that a profile can raise, lower or switch off per client, under Checks, Severities in the profile. Which findings count scopes the findings to the role being reviewed. The full list of checks for your language pair, with each one's current severity and a line on what it does, is in the profile editor and in Help inside the app.

Severity, in plain words​

  • Critical: must fix; one critical finding fails the file whatever the score.
  • Major: look at it; it counts toward the score.
  • Minor: for your awareness; no weight unless the workspace gives minor findings one.

A rule finding the AI disputed stays in the list with its label and counts at half weight until you decide. The AI check of rule findings.