Human review, built
into the process.

The Text QA view presents each chapter as numbered sentence segments — not a wall of text. Every sentence in the translation is individually addressable, so reviewers can pinpoint exactly what needs attention without losing context.

The chapter navigation panel on the left lets reviewers jump directly to any chapter in the book. The active chapter is highlighted and a progress indicator shows which chapters have been reviewed.

Clicking any sentence segment opens the Flag Issue panel. The reviewer sees the exact text being flagged, selects an issue category (Grammar, Accuracy, Terminology, Style, Fluency, and more), writes a comment, and — crucially — can provide a suggested replacement text.

Suggested fixes aren't just notes — they're structured replacements that can be applied directly to the translation file. This closes the loop between human review and the published edition.

This is particularly important for technical vocabulary — AI model names, acronyms, proper nouns, and borrowed foreign words all have pronunciation edge cases that text-to-speech systems get wrong in predictable ways. Our pronunciation dictionary flags high-risk words before they ever reach the reviewer's ear, and improves with every book we complete.

Pronunciation data is one of the clearest examples of our data science loop at work — every reviewer correction refines the dictionary, every dictionary improvement benefits every future edition in that language.