skills/human-writing/references/narrative-choices.md

# Narrative-choice audit

Use this reference for fiction, memoir, anecdotes, case studies, and other prose with characters and events. Apply it before sentence-level polishing.

StoryScope analysed roughly 5,000-word fiction. Its findings describe group-level tendencies, not rules for individual writers. Do not turn the findings into a recipe, force a nonlinear timeline, insert a subplot, or add named cultural references simply to look human. Choose what serves the story.

## Start with intent

State what experience the story is trying to create. Then inspect the narrative across ten lenses: agents, relationships, events, plot, setting, revelation, situatedness and intertextuality, time, perspective, and style. Concentrate on the two or three choices that most affect this story.

## High-value questions

### Theme and moral pressure

- Does the narrator explain the theme after the scenes already conveyed it?
- Do dialogue and subplot exist mainly to restate one central message?
- Is the protagonist's moral position too clean for the situation?

Prefer earned implication over a closing lesson. Keep explicit commentary when it belongs to the narrator's voice, genre, or rhetorical purpose. Ambiguity should arise from competing motives and consequences, not from deleting necessary information.

### Character agency and resolution

- Does every important event serve the protagonist's growth?
- Does the ending resolve through a sudden internal understanding because that is the easiest closure?
- Are supporting characters people with pressures of their own, or devices for the protagonist?

Give characters specific, sometimes conflicting motives. Let consequences come from the interaction of choices, institutions, chance, and other people. A protagonist-driven ending is not a flaw when the story has earned it.

### Plot, causality, and subplots

- Is the causal chain implausibly smooth, with each event neatly preparing the next?
- Does the story contain only one track because every side thread was optimized away?
- Are all loose ends tied, even those whose uncertainty could remain productive?

Consider friction: incomplete knowledge, delayed consequences, competing agendas, or a thematically relevant side thread. Do not add mess for its own sake. Clarity and linearity may be exactly right for a short anecdote, thriller, folktale, or children's story.

### Emotion, body, and interiority

- Does every emotion appear as a tightened chest, caught breath, clenched jaw, cold skin, or environmental metaphor?
- Does the prose avoid ordinary emotion words even when a person would simply think "I was scared"?
- Does deep interior access explain feelings the character has not yet understood?

Vary the channel: direct labels, behavior, dialogue, silence, thought, bodily sensation, or ambiguity. Use the body and senses selectively. One precise physical detail is stronger than a stack of generic reactions.

### Setting and sensory detail

- Does the opening mechanically establish location before anything happens?
- Does the environment mirror the character's mood too perfectly?
- Are smell, weather, light, and spatial detail present because the scene needs them or because description feels mandatory?

Select details through viewpoint and action. Let some settings resist the character's mood. Use exact real-world or cultural detail only when the user or source supplies it or when invention is authorized.

### Time and revelation

- Is the story chronological simply because that was the easiest order to generate?
- Would a time jump, omission, memory, or shifted opening change what the reader knows at a useful moment?
- Does a surprise merely add information, or does it reframe an earlier scene?

Choose chronology deliberately. Nonlinearity is valuable when it controls knowledge, emotion, comparison, or suspense. A gimmicky flashback is worse than a clear linear sequence.

### Perspective and reader relationship

- Is viewpoint consistent by default rather than chosen for effect?
- Does narration give every scene the same distance and emotional temperature?
- Is there a natural place for an aside, omission, direct address, or change in focal distance?

Decide who can perceive, interpret, and withhold each detail. Direct reader address and fourth-wall play are options, not requirements. Use them only when the voice supports them.

### Intertextual and social situatedness

- Does the story rely on vague mythic echoes instead of a lived cultural context?
- Are institutions, work, class, technology, family structures, and public life absent from a story that should involve them?
- Would a supplied, specific reference reveal character or setting better than a generic allusion?

Ground the story in a world larger than the protagonist, but never invent factual references or use cultural specificity as decoration.

### Narrative diversity

StoryScope found that human stories occupied a broader narrative space than stories from the models it tested. The useful lesson is not "add human markers." It is to resist the first statistically familiar solution.

When planning or substantially revising fiction, briefly generate two or three genuinely different choices for the weakest structural element, such as the source of conflict, the timing of a revelation, the moral framing, or the mode of resolution. Select the choice that best fits this story and the writer's intent. Do not expose discarded alternatives unless the user asks.

## Structural revision pass

1. Extract a compact map: central characters and motives, event chain, conflict, resolution, timeline, viewpoint, revelation, setting function, and stated or implied theme.
2. Identify the choices that feel automatic, over-explained, over-tidy, or disconnected from the piece's intent.
3. Revise the smallest set of structural choices that materially improves the reading experience.
4. Check continuity and causality after reordering or changing events.
5. Polish the prose only after the structure holds.

## Do not overfit the research

- Do not treat StoryScope as an authorship oracle; its distributions overlap.
- Do not apply fiction findings to ordinary nonfiction.
- Do not maximize every human-associated feature. That produces another formula.
- Do not sacrifice genre expectations, accessibility, or the writer's voice for statistical rarity.
- Do not claim that a rewrite is undetectable or human-authored.

Adapted from the MIT-licensed [StoryScope repository](https://github.com/jenna-russell/storyscope) and the CC0 paper ["StoryScope: Investigating idiosyncrasies in AI fiction"](https://arxiv.org/abs/2604.03136).