Localization Checker Manual, part 1: the checks and their results

This part explains scan inputs, individual checks, Play mode patrols and result interpretation. The Quick Start provides a walkthrough. Manual part 2 (Manual-Safety.md) covers project safety, prerequisites and limitations, including false positives and missed cases.

How a scan works

For each configured resolution and language:
  copy the scoped Canvases into a hidden preview scene
  remove project scripts and use in-memory font and material copies
  apply all translations for that language together
  rebuild the layout and measure with the text engine
    ├─ text against its box: Overflow
    ├─ layouts that size text: Layout
    ├─ intersecting text bounds: Overlapping texts
    └─ Japanese, Chinese, Korean and Thai: Line breaks
Without a layout pass:
  ├─ look up characters through each font's fallback chain: Missing glyphs
  └─ compare every table key: Placeholders, Rich text tags, Translation

The overview groups these results into five columns: Overflow includes overflow, layout, overlaps and line breaks; Missing glyphs includes character coverage and font configuration; Placeholders and Rich text tags cover string structure; Translation includes table values, right-to-left and encoding concerns, and text outside localization.

What a scan reads

Choose Scope in the Scope row of Before you scan.

• The checker does not open closed scenes. Opening a scene may run [ExecuteAlways] scripts that change assets. Open the required scene yourself, then use Open scenes.
• Scene scans exclude objects tagged EditorOnly because builds strip them. Prefab scans include them.
• The line below the scope dropdown describes the current input. Scans are unavailable in Play mode.

Text components and elements. The scan reads content, font, bounds and layout from TextMeshPro uGUI and legacy UI Text components. On Unity 6, it also reads UI Toolkit documents as described below.

Languages to check. Select languages found in the translation tables. To check font coverage before translations exist, use Add a language: choose one of sixteen common languages or Enter a code…. Language-code matching is case-insensitive. Pseudo-localization adds a generated test language.

Source language. Choose the language in which the strings were originally written. It provides the comparison values for table checks and the content lookup for components containing source text instead of keys. Checks for text outside localization also require it. If it is not selected for checking, the scanner lays it out only as a comparison baseline.

Resolutions. Each configured resolution is measured separately. Overflow results identify the resolution used.

UI Toolkit (Unity 6)

On Unity 6, the scan builds scoped UIDocuments from their UXML in memory, including nested templates. It checks Labels, Buttons and control labels for localization, missing glyphs, overflow, overlaps and layout, plus line breaks from Unity 6000.3. The document copy renders into a checker-owned texture rather than the Game view. This support is unavailable on Unity 2022.3 and Tuanjie Engine.

• Results identify an element by object and element path, such as Menu/Document › panel/title.
• A Unity Localization LocalizedString binding on an element’s text property in UXML gives High binding confidence. The scan cannot see bindings assigned through SetBinding, strings assigned by code, or elements created at runtime.
• Elements hidden by UXML (display: none) are revealed on the copy one group at a time. Their results include hidden in the document.
• From Unity 6000.5, an element with no font in its styles uses the theme’s default font. The engine does not expose that font, so its characters are skipped and the summary counts these elements.
• UI Toolkit layout requires editor frames to advance. Do not add -quit to command-line scans.

Where translations come from

Unity Localization

The checker automatically reads Unity Localization 1.5.2 or later, below 2.0:

• All String Table Collections.
• LocalizeStringEvent bindings whose enabled dynamic-string update listener targets the text component, provided the localization component is also enabled.
• GameObjectLocalizer bindings that track the text component with a String Table. The checker reads per-language fonts from these bindings and custom font components derived from LocalizedAssetEvent. It resolves asset-table fonts by GUID without loading them through Addressables.
• Supported font-size, size, position, enabled-state and font changes from GameObjectLocalizer, applied to the copy before measuring each language.

Translation files

Add files under Translation files (CSV, Excel, JSON, XLIFF, PO), or drag them from the Project window into that group. The file extension selects Format.

• Column mapping. The first column defaults to the key. Headers such as en, English, Deutsch and 日本語 map to language codes. Map unrecognized columns yourself; ID and No are never treated as languages. Edit… opens the mapping.
• Invalid files. A syntax error, duplicate key or missing mapped header prevents the whole file from being read. Its status gives the reason and line number, or worksheet row number.
• i18next. Each language requires item_other and the plural forms its rules use for whole numbers: for example, German item_one, or Russian and Polish item_one, item_few, item_many. A required form that is absent produces Missing translation.
• XLIFF. Inline codes such as <g>, <x/> and <ph>, or units divided into several segments, prevent the file from being read.
• Excel. The parser reads saved cell values and cached formula results, without evaluating formulas or applying cell display formats. Rich-text runs are joined and their formatting discarded. Whole numbers of up to 15 digits are normalized to digits; other numeric values retain their stored representation, so dates usually appear as serial numbers. Save keys and translations as text. An open workbook is read from its last saved version. Legacy .xls and password-protected workbooks are unsupported; save an unprotected .xlsx copy.
• PO entries. Context prefixes the key with a bar, as in menu|Open. An empty msgstr or fuzzy entry produces Missing translation, since gettext falls back to msgid. Plural entries create keys such as item[0], item[1] and so on. Plural-Forms determines the count, defaulting to two; an empty required msgstr[n] is missing, but forms unused by that language are not.
• PO placeholder comparison. With msgid language set, every form is compared with msgid_plural, following msgfmt’s approach. A form used for fewer than five values from 0 through 1000, according to plural=, may omit the number placeholder but cannot add one; other forms must retain all placeholders. English’s singular form is one example. Files must be UTF-8. An unreadable Plural-Forms expression, or one producing an out-of-range form for a number from 0 through 1000, prevents the file from being read.

I2 Localization

The checker automatically reads I2 Localization 2.8.9 or later without calling its API during scans.

• Tables: the I2Languages source in a Resources folder and LanguageSource components in scanned files. Only Text terms become keys.
• Bindings: an enabled Localize component binds the text component on its object to its term with High confidence. A component with no term is Unbound.
• Measurement: the Localize case modifier, prefix and suffix are applied before layout.
• Play mode: the patrol uses I2 to switch the game’s language.

How a text is bound to a key

Key binding describes the match and its confidence.

A key found in several tables reduces High confidence to Medium. Unbound means no key was identified. Broken means the binding names a table or key absent from the sources read. If component contents resemble a table’s keys but the table option is off, the scan offers Turn it on and scan again.

Unreadable tables and duplicate keys belong to Sources. When any exist, the overview shows Problems with the translation tables themselves (a table not read, a key twice): below its counts. Click the count for details.

The checks

Overflow

• Height is checked; width is checked only without wrapping. An axis controlled by Content Size Fitter or a Layout Group is handled by the layout check instead.
• Measurement requires a translation for the bound key in that language. Other reasons to skip include UI text outside a Canvas, custom text or layout classes, input-field labels and dropdown captions rewritten by their controls, a Canvas that could not be copied, missing TMP Essential Resources, or translations referencing font, style, material or sprite assets.
• A Mesh Renderer-based 3D TextMeshPro under a Canvas is measured with it. Outside a Canvas, it is copied with its ancestor RectTransforms and measured in its own box, identically at every resolution. It receives no layout or overlap check and no side-by-side image.
• Screen Space - Camera Canvases use the camera viewport configured in the editor at scan time. A Scroll View with automatically hidden scrollbars reserves scrollbar space only when content overflows.

Common Detail notes:

• hidden in the scene: the component or an ancestor is inactive. Measurement revealed all hidden objects in that Canvas together.
• measured without these scripts: the listed scripts were removed from the copy and may change content, fonts or sizes at runtime.
• the source text too, not grown in this language: for reference: the source has the same issue at that resolution and the translation does not worsen it. A growth note such as the source text too, 20 more in this language means the issue still counts.
• Fixes: for overflow and truncation, the checker tries wrapping, Auto Size or Best Fit, and font sizes down to three quarters of the configured size, on a copy. Examples include fits at font size 30 (now 36) and does not fit wrapped. These are suggestions; no fix is applied to the original component.

Layout

The scan follows each text component’s sizing layout up to its root.

Detail gives distances in canvas units, such as past its parent: left 12, right 12. As with overflow, an unchanged issue already present in the source language is listed for reference only.

Overlapping texts

• Comparisons are limited to text components in the same Canvas that can appear together. Components in a closed dialog are compared only with each other.
• The engine’s rendered text bounds form rectangles. An intersection must be at least 1 unit on each axis. Pairs already overlapping in the original scene are excluded.
• A component with an overflow issue in that language is excluded from overlap comparisons; overflow takes priority.
• Detail identifies the other component and intersection depth, for example covers the glyphs of Canvas/Options/Value, 12 units deep.

Line breaks

• Only automatic breaks are checked. Explicit line breaks in translations are excluded. Japanese and Chinese rules restrict punctuation at line starts and ends and keep pairs such as …… and —— together.
• Thai checks detect syllable violations: a leading vowel (เ แ โ ใ ไ) cannot end a line, and a dependent vowel, tone mark or sign cannot start one. Word-boundary detection requires a dictionary that the checker does not provide. An unmarked break between Thai letters therefore produces May break inside a word, even if it happens to be valid. Detail may say line 2 breaks between มก, where no word boundary is marked. A zero width space (U+200B) marks a boundary for TextMeshPro and UI Toolkit, suppressing that warning at the boundary; legacy Text ignores it.
• Detail lists the break locations, such as line 1 ends with （ · line 3 starts with 。. Hover over the issue kind for guidance.
• TextMeshPro uses the two Line Breaking for Asian languages files in TMP Settings. Missing or empty files provide no rules. TextMeshPro in Unity 6 can still leave opening brackets and quotation marks at line ends. Legacy Text does not apply these rules. UI Toolkit handles most itself; from Unity 6000.3 the scan checks its resulting lines, excluding strings containing rich-text tags or escapes and counting those exclusions in the summary.
• Modern Korean breaks at word spaces. Enabling Use Modern Hangul Line Breaking Rules changes TextMeshPro’s line-breaking behavior only. Legacy Text already breaks at spaces. UI Toolkit has no equivalent switch and can split between syllables; from Unity 6000.3 the scan reports these element-level breaks as Bad line break, for example the word at 니다 split at the end of line 2.

Pseudo-localization

Scan Pseudo-localization generates qps-ploc from source strings, accents letters and expands length by about 40%: Settings becomes ŚéţţîñĝśŚéţţ. Tags and placeholders remain unchanged. This requires no translated table and checks only overflow, overlaps and layout, with Static Estimate confidence.

This generated scan language is separate from the game-provided Pseudo-language used by the Play mode patrol.

Variable samples

Under Variable samples, assign the longest expected runtime value to each placeholder. The scanner substitutes these values when measuring overflow and layout in every language. Use the exact placeholder spelling, such as {playerName}, {0}, %s or %1$s; names are case-sensitive. Unsampled placeholders are measured literally, and Detail identifies them.

In the patrol, Use variable samples asks the game to display the samples. Matching Unity Localization global string variables are set automatically. For custom code, declare a static void Method(IReadOnlyDictionary<string, string> samples) under Variable samples for patrols (the project’s own code). The checker passes null when the original values should be restored.

Missing glyphs

The scan resolves characters through the text system’s font chain without querying the editor computer’s system fonts:

• TextMeshPro: the font asset and its fallbacks, TMP Settings fallbacks, then the default font. Dynamic assets also use their source font file. Bold and italic lookups start with the weight table’s typeface.
• Legacy Text: the assigned font, followed by fonts referenced in its import settings.
• UI Toolkit: the resolved font and its fallbacks, followed by the owning panel’s Text Settings. Panels sharing a font retain separate fallback settings.

Tags and placeholders are excluded. Case styles are applied before lookup: uppercase é, for example, is checked as É.

Language coverage. The checker looks up built-in character sets and punctuation before translations exist. These cover the languages fully supported by Steam, including Chinese, Japanese, Korean, Thai, Arabic, Greek, Russian, Ukrainian, Bulgarian and Latin-script languages. Language coverage reports none, partial with a percentage, complete, or left to device fonts. Font lacks the language is Critical if no language characters are covered, reduced to Warning when system fonts may supply them. Partial dependence on system fonts is Warning; incomplete character or punctuation coverage is Info.

UI Toolkit coverage uses the font’s own fallback chain without a panel’s global fallbacks. An element may therefore render successfully through its panel even when the font’s coverage is incomplete.

Fonts by language lets you provide replacements when your game’s font-switching mechanism is not detected. The scan uses the replacement for character lookup and overflow in that language. Also check the language coverage of every TextMeshPro font asset in Assets adds project-wide font-coverage hints.

The scan’s missing characters by font groups missing characters by font. Copy for Font Asset Creator copies them for Font Asset Creator’s Custom Characters input.

Placeholders

Supported placeholders include {0}, {name}, Smart String formats such as {count:plural:item|items}, %s, %d, %1$s, {[NAME]} and i18next {{name}}. Translations may reorder them.

Rich text tags

The checker interprets tags according to the relevant text engine and compares them with the source string. Intentional source formatting is accepted, as are complete tag pairs added or removed together.

Tag problems are grouped into one issue per key and language. Attribute-only differences are Warning; other problems are Critical. A translated tag value such as <color=红色> is Critical.

Tag names a missing asset is Critical when a TextMeshPro <font="…">, <sprite="…">, <sprite name="…"> or <style="…"> references a name absent from the project. Such references can render incorrectly or appear literally. The issue appears once per name and table under (no language); it does not validate runtime asset resolution.

Translation

These checks require Source language and examine every key in every table, including keys unused by the scanned UI.

Texts outside localization:

A component matched only by its source content is also hard-coded: the match does not establish runtime replacement. May be set by a script means a script field references the text component, or the UIDocument containing the element. Empty values, New Text and strings without letters are excluded.

Ignore list. Key rules suppress eligible per-key table-value issues. They do not suppress UI overflow or glyph issues, or table-level issues such as missing languages, suspected mojibake and missing asset references. Path rules and text regular expressions suppress Hard-coded text, Unbound and May be set by a script, plus the patrol’s Strings joined at run time and Hard-coded at run time. They do not suppress Broken binding or other checks. Ignore adds an applicable rule from the selected issue.

The patrol in Play mode

The patrol asks the game’s localization system to switch languages, then reads the UI as the game laid it out. It does not assign translations to text components or read translation tables. It does not run the table placeholder, rich-text or translation checks, or the scan’s layout checks. Its optional pseudo-language check is described below.

Eligible components and elements. The patrol reads enabled, active TextMeshPro and legacy UI Text components under enabled Canvases in loaded scenes and DontDestroyOnLoad. It excludes those fully hidden by a Canvas Group or culled by RectMask2D. On Unity 6, it also reads displayed elements in active UIDocuments. It reads 3D TextMeshPro when an enabled game camera can render it on screen: the renderer must be enabled, within that camera’s view and visible layer mask, and the camera must not target a texture. These 3D components are framed using that camera and excluded from overlap comparisons. UI text components outside a Canvas are unsupported.

Language switching, shown under Switch in the patrol bar:

• Unity Localization: sets the selected locale through Localization Settings in Play mode and waits for displayed LocalizeStringEvents to load.
• Your own system: invokes the two static methods declared under Language switch for patrols (the project’s own localization system). Returning from the switch method marks the switch complete.
• I2 Localization: becomes available when I2 is installed and switches without saving I2’s language preference.

Per-language checks:

Patrol issues have Play Verified confidence. Overlap comparisons and source-language equivalence require Languages to check to include the source language so the patrol can read that version of the page.

Checking a page. Click Check page in all languages (F8), or focus Localization Checker and press F8. This starts a patrol if needed. For each selected language, it switches, waits Page check: frames to wait after each switch (120 by default), also waits for Unity Localization’s displayed strings, then reads the UI and captures the Game view. It restores the previous language and retains the completed page for this editor session.

Stopping the patrol or leaving Play mode during a check discards that unfinished page; earlier completed pages remain. If an initially visible root Canvas or UI Toolkit document disappears, the patrol ends and explains why. Check again when the screen is stable. An additional popup appearing does not by itself count as leaving the page.

Page names. The suggested name uses “scene: Canvas”, based on the topmost root Canvas and its scene. A DontDestroyOnLoad Canvas uses the active scene; a UI Toolkit-only page uses the first document. Automatically suggested names gain a number when already used, such as Title: MainCanvas (2). A manually entered existing name replaces that page; the bar warns before the check.

Results and persistence. Each completed page has an overview block with language rows and Open screenshot. A dash means not checked, rather than zero issues. In Issues, patrol rows follow scan rows of the same severity and include Page. Map a key… and Scene preview are unavailable for patrol rows. Ignore applies only to the two pseudo-language issue kinds. Completed pages survive domain reloads and leaving Play mode, until cleared or the Editor exits; they are not project assets.

Screenshots. Each language captures the Game view in the frame read, including outlines if Outlines is enabled. Files are kept in Temp/LocalizationChecker/Patrol. Export screenshots… copies them to a chosen location with one folder per language.

Pseudo-language

Your game’s localization system must supply a pseudo-language that brackets each localized string, such as [Šţáŕţ]. The checker does not generate this runtime language. Joined strings produce several bracketed groups in one component; hard-coded words appear outside the groups. Check page in all languages reads this language last and runs only these checks:

• Tags are stripped before checking. Digits, punctuation and spaces outside the brackets, such as scores and times, are accepted. Placeholder values already enclosed within a localized string are accepted too.
• Unity Localization: uses the first Pseudo-Locale in the project. Its final method must be the bracket-adding Encapsulator. Earlier padding may contain bracket characters only if a subsequent Accenter replaces them. An unsuitable configuration skips this check and explains why in the page block.
• I2 Localization and custom systems: enter an implemented pseudo-language code and its brackets under ⚙ Settings › Patrols › Pseudo-language. The default brackets are [ and ]. That code is removed from the normal checked languages and read last. Never assign a real language code here.
• The Pseudo row contributes only to Translation. It does not check overflow, glyphs or overlaps, and TextMeshPro missing-character warnings are not collected while it is active.
• Intentional literal values, such as player names or language names, can be reported. Ignore creates an object-path rule for these issues. Existing patrol results immediately follow the updated ignore list; scans use it on the next run. Text rules match the value displayed in the pseudo-language.

Superseded by the patrol

A retained page can supersede scan issues for the same component and language. The scan row remains for reference, with Detail beginning superseded by the patrol’s page ”…”, checked ….

• Overflow issues require the Game view to match the scan resolution exactly and the displayed string to equal the string measured by the scan.
• Missing glyphs, Left to the device’s fonts and Drawn by a fallback font require a non-truncated TextMeshPro component displaying every character named by the scan issue.
• Layout, encoding, font-configuration, table and unlocalized-text issues are never superseded. Neither are legacy Text character issues, overflow issues with a different runtime string, or issues from prefab assets: runtime instances are separate objects.

If a page captured the previous language before the UI updated, increase the wait frames and check it again. Clear patrol results restores superseded scan issues to the counts.

Characters TextMeshPro reported missing

During a patrol, The patrol’s missing characters by font collects missing-character warnings from any TextMeshPro component the game displays, including player names, chat and server strings. It is not limited to checked pages. Collection depends on TextMeshPro’s Console warnings; Disable warnings in TMP Settings prevents it. The list remains until the next patrol starts.

The command line (continuous integration)

A build server can scan a prefab folder without opening the checker window:

Unity -batchmode -projectPath <project> -executeMethod Fastzhuzhu.LocalizationChecker.Editor.Ugui.CommandLine.Scan -logFile <log>
      [-locchkFolder Assets/UI] [-locchkReport <file.csv>] [-locchkJUnit <file.xml>] [-locchkFailOn critical|warning]

• Scope: all prefabs in -locchkFolder and its subfolders, defaulting to Assets. This matches Prefabs in a folder and opens no scene.
• Settings: uses ProjectSettings/LocalizationChecker.json. Configure the project in the window and commit that file for CI.
• Reports: -locchkReport writes the Export CSV… format. -locchkJUnit writes JUnit XML with one test case per overview cell and failures determined by the run’s verdict. Paths are relative to the project and cannot be under Assets, Packages or a package folder.
• Log: includes the summary, notes and verdict, for example Failed: counted issues at Critical or worse: 3.
• Exit codes: 0 means every prefab was read and no counted issue met or exceeded the -locchkFailOn threshold (critical by default). Reference-only issues do not fail the run. 2 means the threshold was reached or a prefab could not be read. 3 means the run could not complete the request: invalid or repeated arguments, refused report paths, missing language selection, invalid folders, interrupted scans, report-writing failures or unexpected errors. The log gives the reason. 1 indicates a Unity failure, such as compilation errors.
• Do not add -quit. The scan exits the Editor itself; an early exit prevents checks that need editor frames from completing.

Code and data

Code & data finds potential strings outside scene and prefab scans, plus culture-sensitive code calls. These text searches have Low confidence: they do not establish that a string reaches players or a call is incorrect. Both scans are read-only.

• Scan code reads C# under Assets that is included in the game, excluding Editor folders, editor-only assemblies and tests. It lists string values assigned to text or passed to SetText, directly or through string.Format, such as Assets/Scripts/Shop.cs:42 and $"Coins: {coins}". Empty values, the default New Text, strings without letters and lines containing locchk:ignore are excluded. Click a row to highlight the script; double-click to open its line.
• Scan code also lists culture-sensitive calls in a separate card: parameterless ToUpper() and ToLower(); char.ToUpper(c) and char.ToLower(c); float, double and decimal Parse or TryParse without an IFormatProvider; and single-argument Convert.ToSingle, ToDouble and ToDecimal. Turkish casing changes i to İ; German and French numeric conventions can change how 1.5 is parsed. For culture-independent keys, filenames, saves and settings, use ToUpperInvariant / ToLowerInvariant or pass CultureInfo.InvariantCulture. Custom methods with the same names can also be listed. locchk:ignore excludes a line.
• Scan data reads saved ScriptableObject .asset text without loading assets or executing their code. It lists likely human-readable values with at least two words, or at least two characters in scripts such as Chinese, Japanese, Korean or Thai, for example Assets/Data/Shop/Gems.asset › m_ItemName. Single words, LocalizedString fields and I2 sources are excluded. Binary assets cannot be read; use Asset Serialization Force Text. Click to highlight the asset in the Project window, or double-click to select it.

Exports

The Export menu beside Rows listed exports the currently displayed rows in their current order. The file dialog remembers the last export folder. Exports cannot be saved under Assets, Packages or a package folder.

• Export HTML… creates one offline UTF-8 file that follows the system’s light or dark appearance. Click counts to filter by severity, check or language; press / to search; expand rows to compare source and translation. Select in Unity copies a selection command. Return to Unity with Localization Checker open to select the object; objects in closed scenes are pinged without opening the scene. The report describes the exported rows and does not certify that all checks passed.
• Export CSV… writes the same rows as UTF-8 CSV, including resolution, page, check time and object ID. Filter from the overview or enable Hide rows for reference to narrow the export. To prevent spreadsheet formulas, cells beginning with =, +, -, @, a tab or a carriage return receive a leading apostrophe. The same protection applies after semicolons within a cell.
• Export screenshots… copies patrol screenshots into one folder per language.
• Export context package… writes retained patrol pages as context.json and screenshots in your chosen folder. Each nonempty text value has an entry per language and page, including the page, screenshot, text bounds on that image, object, binding keys from open scenes, displayed value, rendered font and size, component dimensions, wrapping and fit. Pseudo-language results and empty labels, such as those on icon-only buttons, are excluded. Missing screenshot files are not recreated. Leave Play mode before exporting because bindings are read from open scenes. format and version identify the JSON schema.
• Export images…, in the side-by-side window, saves PNG files named by object, resolution and language. Invalid filename characters become underscores, long names are shortened, and collisions gain a number to preserve every image.

Reading the results

The window

Top band: logo · Localization Checker · version · author's signature
Toolbar:  ▶ Scan · setup status · ⚙ Settings · window language · ⋮ menu
Tabs:     Overview · Issues · Code & data
Drawer:   Key bindings · Measuring · Ignore list · Patrols
Footer:   summary · scan notes · messages

The ⋮ menu provides Clear scan results, Clear patrol results and About. The overview begins with Before you scan, renamed This scan after scanning. Its rows are Translation sources, Languages to check, Source language, Scope, Scan and Patrols (optional). The first five contain their related settings. Each numbered indicator shows status: green for complete, red for a blocker, yellow for review, and blue for ready. The blue-highlighted row is the first blocking setup step, or Scan when ready. After a scan, rows collapse; editing a setting changes the scan status to Out of date: settings changed. In Play mode, Languages for the patrol replaces this setup list.

Severity and “for reference”

• Critical: high-priority findings with possible player impact. Confirm the context before treating them as defects.
• Warning: potential problems or behavior that may vary by device or context.
• Info: findings worth reviewing or additional diagnostic information.

Some findings are reference-only: Unbound, May be set by a script, fallback-rendered characters, unmeasured input-field labels, unchanged source-language issues and scan issues superseded by patrol pages. Detail identifies these rows. They do not contribute to overview cells or severity totals. Hide rows for reference removes them from the issue list and exports.

Confidence

• Play Verified: observed by the patrol in the running game’s layout.
• Engine Verified: measured by the text engine on a copy.
• Engine Metadata: read from font data.
• Static Estimate: measured using a generated pseudo-translation.
• No value: derived from strings, tables or bindings.

Confidence describes the source of evidence, not whether the finding is an intentional design choice.

The issue list

Default columns are Severity, Issue, Language, Object and Detail. (no language) identifies language-independent issues. Object shows the name followed by its location in lighter text. Retained patrol pages add Page. Right-click the header to enable Check, Confidence, Table, Key and Key binding. Hover over a cell for its full value.

Click a row to select its object. The right-hand details panel shows all fields and applicable actions: Map a key…, Ignore, Side by side and Open screenshot. For patrol issues, the screenshot is the page in that issue’s language.

Two additional Detail notes explain binding caveats:

• mapped by hand, but a component drives it: the mapping is not used.
• a script references the component that drives it: …: runtime code may change the displayed value; check that screen with the patrol.

Since the last scan

When comparable results are replaced, the overview counts New and No longer reported issues. Click either count to filter; Change identifies each row. A disappeared finding may be fixed, ignored or unmeasured, so inspect the cause. Comparisons require matching scope, languages, resolutions, source language and enabled checks. Comparison data is lost when the scan results are cleared or discarded.

Preview, side by side, heat map

• Preview: select a scan overflow issue to temporarily display and outline its translation in the Scene view. Selecting another row or leaving the checker window restores the original. Only components in open scenes saved at least once are supported.
• Side by side: overflow, overlap, line-break and layout issues can render their Canvas or UI Toolkit document at the issue resolution, with one image per language and the source first. The affected component or layout root is outlined in red. Follow selection tracks the selected row, Zoom changes display scale, Render again refreshes, and Export images… saves PNGs. Outline in files includes outlines in exports. Patrol issues instead show the page’s captured screenshots.
• Heat map: the Scene view outlines objects represented by listed issues, using the most severe issue color per object. The legend groups Overflows, Cut short, Missing glyphs, Placeholders and tags, and Nearly overflows, too small, with matching numbers. Reference-only issues, unmeasured text, translation checks and font-asset issues are excluded. Click an outline to select its row.
• Hierarchy marks: in the classic Hierarchy, a colored square marks each object with counted issues. Hover for counts or click to select the corresponding issue.