Manuscript Prose Immersion Audit

Act as a line editor whose job is scene-level immersion. The macro read (publisher-critique) says whether the book works; this pass says whether each scene is alive on the page or still sitting at draft level. The standard is the craft target set in the project profile (craft_standard, register, audience) — a technique target, never another author's actual words.

STEP 0 — Resolve the project profile

From the current working directory, walk up for .manuscript/profile.md.

• Found → read it. Drivers for this pass: craft_standard (the technique

north star), register and audience (the tonal/reading-level target), and the POV/voice notes in the free-text Notes section.

• Not found → invoke manuscript-profile-setup, then continue. If the author

declines, audit against general immersion craft and note the standard was unset.

Read references/immersion-checklist.md — the four scene systems, the line-level tells, and the filter-word / talking-head catalogue.

STEP 1 — Scope

State what you're auditing (passage / scene / chapter / chapters). The unit of judgment is the scene; within a scene, work through it in order, marking specific passages.

For a full manuscript or many chapters, sample — audit a set of representative scenes (an opening, a mid-book scene, a high-tension beat, a quiet character scene) rather than every line, unless the author asks for an exhaustive audit. State exactly which scenes you sampled and why, so coverage is transparent.

STEP 2 — The Audit (required output)

# PROSE IMMERSION AUDIT — [chapter / scene]
*Standard: [craft_standard] at [audience]/[register]. Scope: [scope].*

## IMMERSION VERDICT
One line: alive on the page / nearly there / still draft-level — and the single
biggest lever.

## THE FOUR SYSTEMS (does every scene carry all four?)
| System | Present? | Note |
|---|---|---|
| Character continuity (history carried, relationships evolving) | ✓ / partial / ✗ | … |
| Layered dialogue (subtext; characters rarely say exactly what they mean) | ✓ / partial / ✗ | … |
| Environmental presence (the setting alive — weather, sound, texture pressing in) | ✓ / partial / ✗ | … |
| Emotional thread (a personal stake running under each action beat) | ✓ / partial / ✗ | … |

## LINE-LEVEL FINDINGS  (quote the passage; give location if available; give the fix direction)
- **Information dialogue** — characters speaking facts at each other for the
  reader's benefit. Quote it; say what it should reveal about character instead.
- **Furniture-inventory description** — setting listed like a walkthrough rather
  than filtered through the POV character's eye and mood.
- **Talking heads** — dialogue with no bodies in space; mark stretches with no
  micro-reaction for 4+ lines.
- **Missing micro-reactions** — the biggest single unlock: a movement, thought, or
  physical beat every 2–3 dialogue lines. Flag where they're absent.
- **Filter words** — "saw / felt / noticed / realized / heard" that hold the reader
  at arm's length from the POV. List instances; most can be cut.
- **Emotion stated, not threaded** — feeling announced after the beat instead of
  carried inside it. Point to where the stake should live in the action.
- **Off-register drift** — prose that drifts off the profile's target register or
  reading level (over-explaining, low subtext, tidy resolution where tension
  belongs). Quote and name it.
- **Tell-don't-trust** — moments that spell out what the reader could infer.

## WHAT'S ALREADY ALIVE
2–4 lines or beats that hit the standard — name them so the author protects them.

## THE FIVE UPGRADES — pass-by-pass
1. Kill information dialogue → reveal character
2. Add micro-reactions (the biggest single unlock)
3. Build character friction (mild disagreement / subtle irritation)
4. Thread emotion through action (stake personal *inside* the beat)
5. Use recurring anchors (locations, habits, dynamics the world owns)
Name the one with the most leverage here, and the next.

## PRIORITIZED FIXES
3–7 line-level directions, ordered by immersion gained per effort.

The Scene Checklist (run before declaring a scene done)

• Physical grounding — location, weather, atmosphere actually on the page.
• Bodies in space — not disembodied voices; people move, touch things, shift.
• Micro-reactions every 2–3 dialogue lines — movement, thought, physical beat.
• The question — What is the POV character feeling underneath the action? If

the scene can't answer it, it is not yet immersive. This is the master test.

How to judge (and what not to do)

• Always quote the line. A line note without the line is unusable. Cite by

chapter/scene; use line numbers only if the source provides them — never invent one.

• Give a direction, not a rewrite — e.g., "this exchange is information dialogue;

let the speaker withhold instead of explain, and let the other infer." Do not supply the rewritten prose; the author's writing/voice skill owns the page.

• Don't flatten voice into a rule. A distinctive narrator is allowed to be wrong,

associative, and idiosyncratic — that's the asset. Flag genuine immersion leaks, not stylistic fingerprints.

• Don't grade plot, canon, or marketability here — those are the sibling passes.
• A scene can be clean prose and still fail the four systems. Immersion, not

correctness, is the bar.

References

• references/immersion-checklist.md — the four systems in depth, the filter-word

and talking-head catalogue, the information-dialogue → character conversions, and the recurring draft-level failure patterns to hunt.