Manuscript Management
Uploading a book gets a manuscript into Castwright. The Manuscript tab — one of the six tabs across the top of a book once it's ready (Manuscript / Cast / Voices / Generate / Listen / Log) — is where you keep shaping it afterward: correcting who says what, running a second read for the mistakes only a careful reader catches, and tidying chapters that came in slightly wrong. It's a genuinely ongoing surface, not a one-time step you pass through on the way to generating audio.
The manuscript renders as your book's own prose, paragraph by paragraph, with every sentence carrying a color-coded bar for the speaker Castwright assigned it. Consecutive sentences from the same speaker group into a segment, so a page of dialogue reads as a series of colored blocks rather than a wall of identical text. The narrator gets a neutral grey — deliberately, so narration never competes with the cast for your attention. Below, The Coalfall Commission's Chapter 3: 48 segments, 6 speakers, one line flagged low-confidence, with the chapter sidebar and detected-cast sidebar both visible.
Two ways to correct a speaker assignment, both non-destructive and both logged:
- Drag the boundary. A handle sits between adjacent segments — grab it and slide it to move the line between two speakers, with a peach drop indicator showing where it'll land. On a phone or tablet the same handle answers to touch or pen; there's no separate "mobile way" to do this.
- Select and reassign a span. Highlight any range of text — even mid-sentence — and a popover offers every character in the cast. Castwright splits the sentence at your selection automatically, so you're never forced to reassign a whole line just to fix its second half.
Every reassignment writes an entry to the Log tab, so nothing here is a silent edit — you can always see what changed and when.
- A low-confidence navigator in the header jumps you straight to every line Castwright wasn't sure about (with J/K keys for jumping without touching the mouse) — visible above as "1 low-confidence." This is covered in full on Reviewing Low-Confidence Speaker Tags.
- The chapter sidebar carries a text filter, an amber badge on any chapter with low-confidence lines still open, and a status icon for in-progress, done, or failed chapters — so you can see where attention is still owed without opening every chapter.
- The cast sidebar lets you click a character to filter the manuscript down to just their lines — useful for auditing one character's voice across the whole book — with an inline "Add character" for a speaker Castwright missed entirely.
Beyond fixing lines one at a time, Review Script sends a chapter (or the whole book) to a second LLM pass that reads back over the attribution the way a careful editor would — catching a stray speaker tag, a line split between two people, dialogue buried inside a paragraph of narration, or an emotion that doesn't fit. It surfaces every proposed fix in a diff view you accept or wave off line by line, rather than applying anything silently.
- Detect Emotions re-runs the emotion pass that decides how a line should land (furious, deadpan, broken) across a chapter, without re-running the full analysis.
- Promote first sentence to title fixes a common EPUB import quirk — a chapter's title landing as its first line of narration — lifting it into the heading in one click, shown to you before it commits.
Need to merge or split chapters instead of fixing attribution? That's Restructure, covered in Uploading a Book — it's one click away from the same screen.
Screenshots of the boundary-drag interaction and the Review Script diff view are tracked as a follow-up — both are mid-interaction states the current capture harness doesn't pose yet.
Next: with the cast confirmed and the manuscript in good shape, head to Reviewing Cast & Assigning Voices.