Performance direction

Stage ② Tag is where you say how the book should be performed.

It is optional. A book with no direction narrates perfectly well; direction is what turns competent reading into performance.

Cues

Mark a passage as whispered, urgent, weary, and so on.

You write in a neutral vocabulary and it is compiled for whichever engine you render with. This is the reason direction survives an engine change: the alternative — storing one engine's native syntax — would mean re-tagging the whole book to switch, and, worse, an engine that did not understand the syntax would read the word "whispers" aloud.

Cues are checked against a known list. One that is not on it is discarded rather than passed through, because an unrecognised cue makes delivery flatter than no cue at all.

The cast

Assign lines to characters, and give each character a voice.

This is what keeps a character sounding like themselves in chapter 40 as well as chapter 3 — the thing a human narrator does by memory and a model cannot.

The auto-tagger

✦ Auto-tag proposes direction for a chapter. It reads your manuscript and your story codex, so a character's delivery stays consistent with what you have established about them.

It never returns your prose

This is worth stating plainly, because it is the obvious worry.

The tagger returns structured cue assignments — which passage, which cue, which speaker — keyed by position. Your text is not in its output at all.

So "the tagger rewrote my sentence" is not a defect that gets caught; it is one that cannot occur.

Proposed speakers are checked against your cast, and proposed cues against the known list. Inventions are dropped.

What it will and will not touch

It only adds cues to passages you left alone. Direction you wrote by hand is not overwritten.

Running it costs one flat credit charge per run, quoted on the button.

Direction and a plain narrator

The Tag stage is available whatever engine you have chosen — but on the included narrator the controls explain that it cannot perform cues, and point you at the Setup stage.

This is deliberate. The alternative, hiding the feature until an expressive engine was selected, meant most authors would never discover it existed, because the included narrator is both the default and the only engine on the entry plan.

What to watch for

• Tagging does not re-render anything. Direction is part of a passage's identity, so tagged passages become stale and are re-rendered — and charged — next time you render.
• Tag before you render a whole book. Tagging afterwards means paying for the affected passages twice.
• Pronunciation fixes are separate. They change how words are spoken without changing a character of your manuscript, so they also make passages stale — that is intended, or the fix would never be heard.