#CAT Editor

The CAT Editor is where a machine draft becomes a finished translation. It shows each translation chunk side by side with its source, lets you edit and approve chunk by chunk, and gives you per-chunk tools: quality scores, QA flags, version history, AI improvement, and engine re-runs.

#Opening The Editor

The editor is a full-screen workspace you can open from:

• The Translations screen — the "Open CAT editor" action on any job.
• Home — clicking a recent translation.
• File Translation, after a job completes.

It is deep-linkable per job and chunk, so you can close it and return exactly where you left off.

#Layout

+----------------------+---------------------------------------------+
| Chunk list (sidebar) | Toolbar                                     |
| - navigation         | Source pane   (read-only)                   |
| - review progress    | Target pane   (rich-text editor)            |
| - status per chunk   | Panels: Chunk Override | Version History |  |
|                      |         QA Flags                            |
+----------------------+---------------------------------------------+

• The unit of work is a chunk — a block of text produced by the chunking step, not a single sentence. Chunks that pack several small paragraphs show a "{count} paragraphs" badge on the Source pane.
• The left sidebar lists every chunk in the job with its status, plus a "Review progress" block: approved count, total, percentage, and "All chunks approved" when you finish.
• Both panes offer "View in larger window" to expand into a bigger editing modal.

#Chunk Statuses

Each chunk carries two kinds of status:

Any edit to the target text automatically promotes the chunk from Draft (Machine) to Edited (Human). Approving is explicit — the "Approve & Next" button or Ctrl+Enter.

#Editing And Saving

• Source is read-only. Target is a rich-text editor (placeholder: "Edit translation...").
• The inline toolbar offers Bold, Italic, Underline, Strikethrough, alignment, undo/redo, and subscript/superscript/clear-formatting under "More Options".
• There is no autosave. Click "Save" (or approve) to persist. Each save creates a new entry in Version History, recomputes QA flags, and refreshes the live preview.
• Editor undo/redo (Ctrl+Z/Y) is per-chunk and is lost when you switch chunks. To undo a saved change, use the Version History panel and its "Revert" action.

#Formatting Limits You Should Know

• EPUB jobs run in text-only mode. A banner explains: "EPUB Translation Mode: only text edits are exported. Formatting changes are preserved from the original EPUB." All formatting buttons are disabled — the export re-injects your text into the original EPUB structure, which is what keeps the book's real formatting intact.
• For DOCX/PDF jobs, inline marks (bold/italic/underline) survive export, but block-level structure created in the editor (headings, lists, blockquotes, code blocks) is not exported, so those controls are disabled.
• Tables and images are not rendered inside the editor panes. Verify them in the exported file.

See Editing Without Breaking Formatting for a review workflow built around these constraints.

#Keyboard Shortcuts

The fastest review loop for long documents: read, fix, Ctrl+Enter, repeat. The "Pending" navigation buttons ("Previous unreviewed" / "Next unreviewed") skip chunks you have already approved.

#Quality Scores And QA Flags

• Each chunk shows a quality score badge. The tooltip breaks it down: glossary accuracy, quality percentage, and the formula "Score = (2×Quality + Glossary) / 3". Scores use 0–100 bands: ≥85 Excellent, ≥70 High, ≥50 Medium, ≥30 Low. Very short segments are marked "⚠ Short segment" because their estimates are less reliable.
• Scoring blends heuristics with COMET (an optional neural quality-estimation model, ~1–2 GB, enabled in Settings) and shows BLEU only when you provide a reference translation. A "Recalculate score" toolbar action re-scores the current chunk on demand.
• The "QA Flags" panel lists concrete issues detected on save, for example: "Translation is empty.", "Tag count mismatch…", "Numbers/units appear to differ between source and translation.", "Translation is identical to source text.", and glossary flags like "Glossary term "X" not translated as one of: "t1", "t2"." When everything is clean it shows "No QA issues detected."
• The "Layers" button opens the "Translation Layers Inspector" — a per-chunk trace of every stage that touched the text (first translation layer, AI validation, rewrite, technical check, Improve With AI, chunk overrides) with diffs and a modification timeline.

#Fixing A Bad Chunk — Your Three Tools

A failed chunk shows the banner: "This chunk failed. Type your translation below and Save, or use Re-run chunk with a different engine."

#Finding Chunks Across A Job

Search and filtering live in the Translations screen's job detail view (not inside the editor):

• Status filter (All Statuses, pipeline statuses, and the three editor statuses), text search ("Search source or target text..."), a "Case sensitive" checkbox, and "Show only failed chunks".
• "Find and Replace" performs bulk replacements across chunks — in source or target text, with case sensitivity, a live "{count} chunk(s) match" preview, and "Replace" / "Replace All".