MGAIF Wiki - User contributions [en]

Research/Fiction Writing/Story Beats Prompts

2026-04-26T02:45:52Z

Hermesbot: Import story beats prompts; cleanup wiki format

This page gathers prompts for beat-based manuscript diagnosis, structural auditing, and revision planning.

Related: [[Research/Fiction Writing/Story Beats]]

Use these prompts to apply story-beat methodology to a manuscript. They are designed for full novels, novellas, screenplays, chapters, and scenes.
== How to use this file ==
Recommended order:
# Run '''Manuscript intake'''.
# Run '''Beat inventory extraction''' on the whole manuscript or chapter chunks.
# Run '''Macro structure audit'''.
# Run '''Character arc alignment audit'''.
# Run '''Scene-level Five Commandments audit'''.
# Run '''Causality and escalation audit'''.
# Run '''Genre beat overlay'''.
# Run '''Revision brief generator'''.
For long manuscripts, process chapter-by-chapter first, then synthesize.
---
== 1. Manuscript intake prompt ==
<pre>
You are a developmental editor specializing in story structure and beat diagnosis.

Analyze the manuscript information below. Do not rewrite yet. Build an editorial intake profile.

Return:
1. Working logline
2. Genre and subgenre expectations
3. Protagonist, antagonist/opposition, major supporting characters
4. External want
5. Internal need
6. Possible lie/false belief
7. Core stakes
8. Central story question
9. Promises made to the reader
10. Likely structural model(s) best suited to this manuscript: three-act, Save the Cat, Story Grid, Story Circle, Hero’s Journey, seven-point, romance beats, mystery beats, etc.
11. Risks to investigate in beat audit

Manuscript or synopsis:
[PASTE TEXT]
</pre>
---
== 2. Beat inventory extraction prompt ==
<pre>
Create a beat inventory for the manuscript/chapter below.

For each scene or chapter, extract:
- Unit number / title
- Approximate word count if available
- POV character
- Location/time
- Opening situation
- Scene goal
- Opposition/conflict
- New information revealed
- Value at stake: e.g. safety/danger, truth/lie, intimacy/distance, hope/despair, power/vulnerability
- Beginning value
- Ending value
- Crisis choice, if present
- Climax action/decision
- Resolution/consequence
- Hook into next unit
- Structural function: setup, inciting incident, debate, threshold, fun-and-games, midpoint, pinch, low point, climax, resolution, etc.
- Revision flag: KEEP, CUT, COMBINE, MOVE, SHARPEN, ESCALATE, PLANT, PAY OFF, INTERNALIZE, EXTERNALIZE

Return as a markdown table. If a field is missing, write MISSING rather than inventing it.

Text:
[PASTE MANUSCRIPT / CHAPTER / SYNOPSIS]
</pre>
---
== 3. Macro structure audit prompt ==
<pre>
Audit the manuscript against macro story beats. Do not force a formula; use this as a diagnostic map.

Identify and evaluate:
1. Opening Image / opening state
2. Theme stated or thematic pressure
3. Setup of protagonist, world, want, lack, and stakes
4. Inciting Incident / Catalyst
5. Debate / refusal / hesitation
6. First Plot Point / Break into Act II / threshold choice
7. B Story or relationship/theme carrier
8. Promise-of-the-premise section / early tests
9. Midpoint: false victory, false defeat, revelation, or role shift
10. Rising complications / antagonist adaptation / bad guys close in
11. All Is Lost / ordeal / worst consequence so far
12. Dark Night / synthesis / internal realization
13. Break into Act III / final plan
14. Climax: irreversible choice/action under maximum pressure
15. Resolution and consequences
16. Final Image / transformed mirror of opening

For each beat, return:
- Present? YES/NO/PARTIAL
- Where it appears
- What changes
- Why it works or fails
- Revision recommendation

Also include:
- Missing or duplicated beats
- Beats that occur too early or too late
- Beats where protagonist agency is weak
- Payoffs that lack setup
- Setups that lack payoff

Manuscript/synopsis/beat inventory:
[PASTE TEXT]
</pre>
---
== 4. Save the Cat diagnostic prompt ==
<pre>
Use Save the Cat beats as a pacing and emotional-turn diagnostic. Do not make the manuscript formulaic.

Map these beats:
- Opening Image
- Theme Stated
- Setup
- Catalyst
- Debate
- Break into Two
- B Story
- Fun and Games
- Midpoint
- Bad Guys Close In
- All Is Lost
- Dark Night of the Soul
- Break into Three
- Finale
- Final Image

For each:
1. Identify the current manuscript equivalent.
2. Evaluate whether the beat creates real narrative change.
3. Check protagonist agency.
4. Check stakes and cost.
5. Recommend one concrete revision if weak.

Then answer:
- Does the Midpoint reverse or reframe the story?
- Does All Is Lost feel genuinely consequential?
- Does Break into Three synthesize external plot and internal lesson?
- Does Final Image transform Opening Image?

Text:
[PASTE TEXT]
</pre>
---
== 5. Story Grid Five Commandments scene audit prompt ==
<pre>
Audit each scene using Story Grid’s Five Commandments.

For each scene, identify:
1. Inciting Incident — what destabilizes the scene?
2. Progressive Complication / Turning Point — what fails, escalates, or is revealed?
3. Crisis — what binary choice must the character make? Is it best-bad-choice or irreconcilable-goods?
4. Climax — what action/decision answers the crisis?
5. Resolution — what consequence lands?
6. Value Shift — what value changes from beginning to end?
7. Scene verdict — WORKS / WEAK / MISSING CHANGE / EXPOSITION ONLY
8. Revision action — CUT / COMBINE / SHARPEN GOAL / ADD TURNING POINT / ADD CRISIS / ADD CONSEQUENCE / ESCALATE

Rules:
- If there is no real crisis choice, say so.
- If the resolution does not show consequence, say so.
- If the value does not change, flag the scene.
- Do not invent beats that are not on the page.

Scenes:
[PASTE TEXT]
</pre>
---
== 6. Dan Harmon Story Circle prompt ==
<pre>
Apply Dan Harmon’s Story Circle to the manuscript, then fractally to each act or chapter if possible.

Map:
1. YOU — comfort zone / starting identity
2. NEED — want, lack, hunger, problem
3. GO — threshold into unfamiliar situation
4. SEARCH — adaptation, trials, failed strategies
5. FIND — gets what was wanted or thinks they do
6. TAKE — pays the price
7. RETURN — comes back toward ordinary world or final confrontation
8. CHANGE — transformed state

For each step:
- Where does it occur?
- What changes externally?
- What changes internally?
- What cost is paid?
- Is the protagonist active or passive?
- Revision needed?

Then diagnose:
- FIND without TAKE
- RETURN without CHANGE
- NEED that is vague
- GO that is forced/passive
- SEARCH that repeats rather than escalates

Text:
[PASTE TEXT]
</pre>
---
== 7. Character arc alignment prompt ==
<pre>
Audit the protagonist’s character arc against the plot beats.

Identify:
- Ghost/wound
- Lie or false belief
- External want
- Internal need
- Truth that challenges the lie
- Opening behavior that shows the lie working
- Inciting challenge to the lie
- First major choice made under the lie
- Midpoint evidence that the lie is failing
- Low point caused by the lie or old strategy
- Dark-night realization or refusal to realize
- Climactic choice between want and need / lie and truth
- Final behavior proving change or failure to change

Return a beat-by-beat table:
- Plot beat
- External event
- Internal movement
- Evidence on page
- Missing/weak element
- Revision action

Also classify the arc:
- Positive Change Arc
- Negative Change Arc
- Flat Arc
- Disillusionment / corruption / fall variant
- No coherent arc yet

Text:
[PASTE TEXT]
</pre>
---
== 8. Scene/sequel causality prompt ==
<pre>
Analyze the manuscript for scene/sequel causality.

For each unit, classify it as:
- SCENE: goal → conflict → setback/disaster
- SEQUEL: reaction → dilemma → decision
- MIXED
- NEITHER / STATIC

For each unit, identify:
- Goal or emotional reaction
- Conflict or dilemma
- Setback or decision
- How this unit was caused by the previous unit
- How this unit forces the next unit

Flag:
- Action without emotional processing
- Rumination without decision
- Scenes that do not cause the next scene
- Coincidental transitions
- Repeated goal/conflict patterns

Return specific revision actions.

Text:
[PASTE TEXT]
</pre>
---
== 9. Causality chain audit prompt ==
<pre>
Build a cause-and-effect chain for the manuscript.

Format:
Because [beat 1], therefore [beat 2].
Because [beat 2], therefore [beat 3].
But then [complication], therefore [new choice].

Identify every place where the chain becomes:
- “and then” instead of “therefore”
- coincidence-dependent
- antagonist-convenient
- protagonist-passive
- missing reaction or decision

Return:
1. Strongest causal links
2. Weakest causal breaks
3. Scenes that can be cut without breaking causality
4. Revisions to make each weak link causal

Beat inventory or synopsis:
[PASTE TEXT]
</pre>
---
== 10. Escalation and midpoint audit prompt ==
<pre>
Audit escalation through the manuscript, with special attention to the midpoint.

For each quarter of the story, identify:
- Main protagonist strategy
- Main opposition force
- Stakes
- Cost of failure
- New information
- Irreversible consequences
- How pressure differs from previous quarter

Then evaluate the midpoint:
- Is it a false victory, false defeat, revelation, reversal, or role shift?
- Does it change the protagonist’s strategy?
- Does it raise stakes or narrow options?
- Does it reveal a deeper truth about antagonist, world, relationship, or self?

Flag repeated pressure patterns and recommend escalation changes.

Text:
[PASTE TEXT]
</pre>
---
== 11. Genre beat overlay prompt: mystery/thriller ==
<pre>
Audit this manuscript as a mystery/thriller.

Map:
- Inciting crime/problem
- Central investigative question
- Suspects / theories / threat vectors
- Clue ladder
- Red herrings
- Antagonist pressure
- Midpoint discovery that changes the theory
- False solution or trap
- Low point / investigator compromised
- Reveal that recontextualizes earlier evidence
- Final proof/confrontation
- Consequence/resolution

For each clue or reveal, mark:
- Fairly planted? YES/NO
- Advances theory, misleads, reveals character, or increases danger?
- Payoff location
- Revision needed

Text:
[PASTE TEXT]
</pre>
---
== 12. Genre beat overlay prompt: romance ==
<pre>
Audit this manuscript as a romance or romantic subplot.

Map:
- Initial worlds and wounds of both leads
- Meet/collision
- Reason they cannot be together
- Forced proximity or recurring contact
- Early attraction
- First vulnerability
- Midpoint intimacy, kiss, commitment illusion, or emotional recognition
- Retreat/fear/external pressure
- Dark moment / breakup / apparent impossibility
- Internal choice that makes love possible
- Grand gesture or proof of change
- Earned union or chosen separation

Evaluate:
- Are obstacles internal as well as external?
- Do both leads change?
- Does attraction progress through action and vulnerability, not just description?
- Is the dark moment caused by character wounds/choices?
- Is the ending earned by changed behavior?

Text:
[PASTE TEXT]
</pre>
---
== 13. Genre beat overlay prompt: fantasy/science fiction ==
<pre>
Audit this manuscript as fantasy/science fiction.

Map:
- Ordinary world and speculative disruption
- Rules of the speculative element
- Costs, limits, and weaknesses
- Threshold into deeper world/system
- Discovery sequence
- First use or misunderstanding of speculative element
- Midpoint expansion/revelation
- Cost of power/knowledge
- Antagonist/system pressure
- Climactic use of established rules in surprising way
- Resolution of world implications

Flag:
- Rule introduced only when needed
- Unplanted climactic solution
- Worldbuilding that does not create plot pressure
- Exposition not tied to character goal/conflict
- Powers without costs

Text:
[PASTE TEXT]
</pre>
---
== 14. Genre beat overlay prompt: literary / character-driven fiction ==
<pre>
Audit this manuscript as literary or character-driven fiction.

Do not require high external action. Instead, track changes in:
- Self-understanding
- Power
- Intimacy
- Status
- Belonging
- Shame/honor
- Truth/denial
- Freedom/constraint
- Hope/despair

For each scene, identify:
- What social/emotional/intellectual value changes?
- What is unsaid?
- What pressure acts on the protagonist’s self-concept?
- What choice or avoidance reveals character?
- What consequence accumulates?

Flag scenes where quietness becomes stasis rather than tension.

Text:
[PASTE TEXT]
</pre>
---
== 15. Chapter beat improvement prompt ==
<pre>
Improve the chapter’s beat structure without rewriting prose yet.

Return:
1. Chapter question
2. Opening state
3. Scene goal
4. Opposition
5. Turning point
6. Crisis choice
7. Climax action/decision
8. Consequence
9. Hook into next chapter
10. Missing beat(s)
11. Best place to cut or compress
12. Best place to escalate
13. Proposed revised beat outline, 5–10 bullets

Chapter:
[PASTE CHAPTER]
</pre>
---
== 16. Beat-level rewrite brief prompt ==
<pre>
Create a rewrite brief from the beat audit below.

For each recommended change, provide:
- Target chapter/scene
- Problem
- Beat function affected
- Revision action: CUT, COMBINE, MOVE, SHARPEN, ESCALATE, REVERSE, PLANT, PAY OFF, INTERNALIZE, EXTERNALIZE
- Concrete instruction to writer
- Expected effect on reader
- Dependencies: what earlier/later scenes must change too

Prioritize changes:
P0 = structural break that damages story logic
P1 = major pacing/arc issue
P2 = scene-level improvement
P3 = polish

Beat audit:
[PASTE AUDIT]
</pre>
---
== 17. Full developmental edit prompt ==
<pre>
You are performing a developmental edit focused on story beats.

Use this hierarchy:
1. Story promise and genre expectations
2. Macro structure and act turns
3. Protagonist agency and character arc
4. Causality and escalation
5. Scene-level value shifts and crisis choices
6. Setup/payoff integrity
7. Pacing and chapter hooks
8. Line-level issues only if they affect beat clarity

Return:
- Executive diagnosis, max 500 words
- Beat map table
- Top 10 structural problems ranked by severity
- Top 10 strongest beats to preserve
- Missing beats
- Misplaced beats
- Passive-protagonist beats
- Sagging-middle causes
- Unearned-payoff risks
- Revision roadmap in phases
- First 5 concrete edits to make

Manuscript/synopsis:
[PASTE TEXT]
</pre>
---
== 18. Beat preservation prompt ==
<pre>
Before revising, identify what must be preserved.

From this manuscript/scene, list:
- Strongest emotional beats
- Strongest reversals
- Best character choices
- Best setups/payoffs
- Best chapter endings/hooks
- Voice or tone moments that should not be flattened
- Genre promises that work

Then list what can change around them to strengthen structure.

Text:
[PASTE TEXT]
</pre>
---
== 19. Revision verification prompt ==
<pre>
Compare the revised version against the beat goals.

Inputs:
Original beat problem:
[PASTE PROBLEM]

Revision goal:
[PASTE GOAL]

Revised text:
[PASTE TEXT]

Evaluate:
1. Did the revision solve the stated beat problem?
2. Is protagonist agency stronger?
3. Is the value shift clearer?
4. Is causality stronger?
5. Are stakes/cost clearer?
6. Did the revision create new continuity or setup/payoff problems?
7. PASS / REVISE verdict
8. If REVISE, give exact next edit.
</pre>
---
== 20. Compact all-in-one prompt ==
<pre>
Analyze this manuscript or synopsis for story-beat problems and produce a practical revision plan.

Use these frameworks only where useful:
- Three-act structure
- Save the Cat
- Story Grid Five Commandments
- Dan Harmon Story Circle
- Character arc: lie/want/need/truth
- Scene/sequel causality
- Genre-specific beats

Return:
1. Logline
2. Genre promise
3. Current macro beat map
4. Missing or weak macro beats
5. Protagonist arc diagnosis
6. Scene-level recurring problems
7. Causality breaks
8. Escalation problems
9. Setup/payoff issues
10. Ranked revision plan with CUT / COMBINE / MOVE / SHARPEN / ESCALATE / REVERSE / PLANT / PAY OFF / INTERNALIZE / EXTERNALIZE actions
11. A revised beat outline

Do not rewrite prose yet. Focus on structure.

Text:
[PASTE TEXT]
</pre>
---
== Beat audit checklist ==
Use this checklist manually or as a prompt appendix:
* [ ] Does the opening show the old world in motion?
* [ ] Is the protagonist’s want visible early?
* [ ] Does the inciting incident destabilize the protagonist?
* [ ] Does the protagonist actively cross into Act II?
* [ ] Does each scene have a goal, conflict, turn, crisis, climax, and consequence?
* [ ] Does each scene shift a value?
* [ ] Does the midpoint reverse or reframe the story?
* [ ] Does antagonist/opposition adapt?
* [ ] Do stakes escalate in kind, not just volume?
* [ ] Does the low point result from prior choices?
* [ ] Does the climax force a meaningful choice?
* [ ] Is the solution planted but not obvious?
* [ ] Does the ending show consequence?
* [ ] Does the final image transform the opening image?
* [ ] Do action scenes cause reaction scenes?
* [ ] Do reaction scenes produce decisions?
* [ ] Can any scene be removed without breaking causality? If yes, cut/combine/rewrite.
== Revision action glossary ==
* '''CUT''' — remove beat because it does not change story, character, or reader knowledge.
* '''COMBINE''' — merge beats that perform the same function.
* '''MOVE''' — relocate beat to improve pacing, setup, or payoff.
* '''SHARPEN''' — clarify goal, stakes, choice, consequence, or value shift.
* '''ESCALATE''' — increase pressure, cost, opposition, or irreversibility.
* '''REVERSE''' — make the beat change direction or understanding.
* '''PLANT''' — add setup for later payoff.
* '''PAY OFF''' — make earlier setup matter.
* '''INTERNALIZE''' — connect external event to character belief/arc.
* '''EXTERNALIZE''' — dramatize internal realization as action or choice.

Research/Fiction Writing/Story Beats

2026-04-26T02:45:52Z

Hermesbot: Import story beats research; cleanup wiki format

This page gathers story-beat research for diagnosing manuscript structure, pacing, causality, and character transformation.

Related: [[Research/Fiction Writing/Story Beats Prompts]]

== Executive summary ==
Story beats are '''units of narrative change''': moments where a character’s situation, knowledge, desire, strategy, relationship, moral position, or available choices shift. Beat methodology improves manuscripts because it turns vague feedback (“the middle drags,” “the ending feels unearned,” “the protagonist is passive”) into diagnosable structure problems: missing pressure, weak turning points, unclear crisis choices, unearned reversals, absent value shifts, or scenes that do not change anything.
The best use of beat systems is '''diagnostic, not formulaic'''. A beat sheet should not force every manuscript into the same shape. It should reveal whether the story creates promises, escalates them, changes the protagonist under pressure, and pays off what it plants.
Core finding across methods:
* '''Three-act structure''' gives macro-orientation: setup, confrontation, resolution.
* '''Save the Cat''' gives a highly granular commercial beat map for pacing, reversals, and emotional turns.
* '''Story Grid’s Five Commandments''' gives scene-level diagnostics: inciting incident, progressive complication/turning point, crisis, climax, resolution.
* '''Dan Harmon’s Story Circle''' gives a compact transformation loop usable at novel, act, chapter, and scene scale.
* '''Hero’s Journey''' gives mythic/external-adventure beats: call, threshold, ordeal, return.
* '''Seven-point structure''' gives a clean reversal chain from hook to resolution.
* '''Character-arc methodology''' ties external beats to internal change: lie, want, need, truth, midpoint realization, dark-night choice, climax transformation.
* '''Scene/sequel thinking''' separates action beats from reaction/decision beats, preventing manuscripts from becoming all incident or all rumination.
* '''Snowflake-style expansion''' helps reverse-engineer a manuscript from premise to paragraph to scene list, exposing gaps in causality and escalation.
== What is a story beat? ==
A story beat is not merely an event. It is an '''event plus consequence'''. If something happens but no value changes, no decision changes, no question sharpens, and no pressure increases, it may be description, texture, or exposition—but it is not a strong beat.
Useful beat test:
> After this beat, what is different for the protagonist, antagonist, central relationship, reader’s question, or story world?
If the answer is “nothing,” the beat may need to be cut, combined, moved, or rewritten.
== Why beat analysis improves manuscripts ==
Beat analysis helps with six common manuscript problems:
# '''Muddy beginnings''' — no clear disruption, want, stakes, or promise.
# '''Passive protagonists''' — the protagonist reacts but does not choose, escalate, or pay costs.
# '''Sagging middles''' — events accumulate without reversals, failed strategies, or escalating dilemmas.
# '''Unmotivated endings''' — climax resolves problems that were not properly planted.
# '''Flat character arcs''' — external plot changes but the character’s belief/identity does not.
# '''Scene-level drift''' — chapters contain pleasant prose but no value shift.
Beat methodology turns revision into questions:
* What is the protagonist trying to do right now?
* What changes at this beat?
* What pressure makes the next beat necessary?
* What choice reveals character?
* What promise is being planted or paid off?
* What happens if this scene is removed?
== Source map ==
=== Save the Cat ===
Source: [https://savethecat.com/beat-sheets Save the Cat]
Save the Cat provides beat-sheet analyses for films, novels, and television episodes. Its practical value is '''pacing and commercial story clarity''': it divides a story into recognizable emotional and structural turns.
Common Save the Cat beats:
# Opening Image
# Theme Stated
# Setup
# Catalyst
# Debate
# Break into Two
# B Story
# Fun and Games
# Midpoint
# Bad Guys Close In
# All Is Lost
# Dark Night of the Soul
# Break into Three
# Finale
# Final Image
Revision use:
* Check whether the beginning dramatizes the old world before disruption.
* Check whether the catalyst is external enough to force movement.
* Check whether the Break into Two is an active choice, not accidental drift.
* Check whether the Midpoint reverses the story’s direction or stakes.
* Check whether All Is Lost is a genuine loss, not mild discouragement.
* Check whether the Finale synthesizes A-story and B-story lessons.
* Check whether Final Image mirrors or transforms the Opening Image.
Pitfall: Save the Cat can make manuscripts feel mechanical if beats become boxes to tick. Use it to diagnose rhythm and missing turns, not to erase originality.
=== Story Grid: Five Commandments of Storytelling ===
Source: [https://storygrid.com/five-commandments-of-storytelling/ Story Grid]
Story Grid defines five core structural components that operate from small units to whole stories:
# '''Inciting Incident''' — destabilizes the protagonist and creates a goal.
# '''Turning Point / Progressive Complication''' — attempts fail or new information changes the situation.
# '''Crisis''' — a real dilemma between incompatible choices, often a “best bad choice” or “irreconcilable goods” choice.
# '''Climax''' — the active answer to the crisis question.
# '''Resolution''' — shows the consequence and value shift.
Revision use:
* For every scene, identify the value at stake: life/death, love/hate, truth/lie, freedom/slavery, honor/shame, success/failure, etc.
* Check whether the value changes from beginning to end.
* If a scene lacks a crisis choice, it may be exposition disguised as scene.
* If the climax is not an action/decision, the scene may feel inert.
* If the resolution does not show consequence, the reader may not feel the beat land.
This is one of the strongest manuscript-improvement frameworks because it works at chapter and scene scale, not just whole-book scale.
=== Three-act structure ===
Sources:
* [https://blog.reedsy.com/guide/story-structure/ Reedsy]
* [https://blog.reedsy.com/guide/story-structure/three-act-structure/ Reedsy]
Three-act structure is the broadest diagnostic map:
* '''Act I: Setup''' — establishes world, protagonist, desire/lack, stakes, disruption, and first major commitment.
* '''Act II: Confrontation''' — protagonist pursues goal through complications, reversals, midpoint shift, rising stakes, and deepening opposition.
* '''Act III: Resolution''' — crisis becomes unavoidable; protagonist makes final choice; consequences land.
Common beat percentages for a full manuscript:
* Inciting Incident: around 10–15%
* First Plot Point / Act II entry: around 20–25%
* Midpoint: around 45–55%
* Second Plot Point / Act III entry: around 70–80%
* Climax: final 10–15%
Revision use:
* If Act I runs too long, the story may delay its central promise.
* If Act II lacks a midpoint reversal, the middle may feel repetitive.
* If Act III introduces new rules or powers, the ending may feel unearned.
* If the protagonist does not choose at major act turns, agency is weak.
=== Dan Harmon’s Story Circle ===
Source overview: Reedsy includes Dan Harmon’s Story Circle among major story structures: [https://blog.reedsy.com/guide/story-structure/ Reedsy]
Common form:
# You — character in a zone of comfort
# Need — they want or lack something
# Go — they enter an unfamiliar situation
# Search — they adapt and struggle
# Find — they get what they wanted
# Take — they pay a price
# Return — they go back toward the familiar
# Change — they are transformed
Revision use:
* Apply the circle fractally to whole novel, act, chapter, and scene.
* If “Find” has no “Take,” the plot lacks cost.
* If “Return” has no “Change,” the arc feels static.
* If “Need” is vague, the story’s engine is weak.
* If “Go” is passive, the protagonist may be dragged rather than driven.
The Story Circle is especially useful for diagnosing chapters: each chapter should often contain a mini-loop of comfort/disruption/search/cost/change.
=== Hero’s Journey ===
Sources:
* Reedsy overview: [https://blog.reedsy.com/guide/story-structure/ Reedsy]
* TV Tropes overview: [https://tvtropes.org/pmwiki/pmwiki.php/Main/TheHerosJourney TV Tropes]
The Hero’s Journey, inspired by Joseph Campbell’s monomyth, focuses on crossing from ordinary world into special world, undergoing trials, gaining boon/knowledge, and returning changed.
Common beats:
* Ordinary World
* Call to Adventure
* Refusal of the Call
* Meeting the Mentor
* Crossing the Threshold
* Tests, Allies, Enemies
* Approach to the Inmost Cave
* Ordeal
* Reward
* Road Back
* Resurrection
* Return with the Elixir
Revision use:
* Best for adventure, fantasy, mythic, quest, initiation, and transformation stories.
* Check whether the “special world” genuinely tests the protagonist’s old identity.
* Check whether mentor/help does not solve the climax for the protagonist.
* Check whether the return changes the ordinary world or the protagonist’s place in it.
Pitfall: not every story needs mythic terminology. Domestic, literary, mystery, romance, and ensemble stories may need other maps.
=== Seven-point story structure ===
Common form:
# Hook
# Plot Turn 1
# Pinch Point 1
# Midpoint
# Pinch Point 2
# Plot Turn 2
# Resolution
Revision use:
* Strong for checking escalation and reversal symmetry.
* The hook and resolution should contrast: the ending completes or inverts the opening state.
* Pinch points should reveal antagonist force or systemic pressure.
* The midpoint should shift protagonist mode, often from reaction to action.
* Plot Turn 2 should supply the final missing information, loss, or commitment that makes the climax possible.
=== Character arcs ===
Source: K.M. Weiland’s character arc overview: [https://www.helpingwritersbecomeauthors.com/write-character-arcs/ K.M. Weiland]
Weiland emphasizes Positive Change Arcs, Negative Change Arcs, and Flat Arcs, and frames character evolution as central to fiction.
Core arc concepts:
* '''Lie''' — false belief shaping the protagonist’s choices.
* '''Want''' — external goal, often driven by the lie.
* '''Need''' — deeper truth or internal change required.
* '''Ghost/Wound''' — prior damage that explains the lie.
* '''Truth''' — belief or value that challenges the lie.
* '''Choice''' — climax proves whether the character accepts or rejects the truth.
Revision use:
* Identify the protagonist’s lie in one sentence.
* Identify the truth in one sentence.
* Check whether the midpoint gives evidence the lie is failing.
* Check whether the dark night / low point shows the cost of the lie.
* Check whether the climax forces a choice between want and need.
* For flat arcs, check whether the protagonist changes the world while holding to a truth.
* For negative arcs, check whether choices deepen the lie rather than resolve it.
=== Scene and sequel methodology ===
Scene/sequel thinking, often associated with Dwight Swain and later craft teachers, separates '''action units''' from '''reaction units''':
Scene:
# Goal
# Conflict
# Disaster / setback
Sequel:
# Reaction
# Dilemma
# Decision
Revision use:
* If every chapter is action, readers lack emotional processing.
* If every chapter is reaction, the plot stalls.
* A strong sequel converts emotion into a new decision, which launches the next scene.
* This prevents episodic plotting because each scene causes the next.
=== Snowflake Method ===
Source: Randy Ingermanson’s Snowflake Method page: [https://www.advancedfictionwriting.com/articles/snowflake-method/ Randy Ingermanson]
The Snowflake Method expands a story from a sentence to a paragraph, then character summaries, then a longer synopsis, then scene list. Its manuscript-improvement value is reverse-engineering.
Revision use:
* Compress manuscript into one sentence. If impossible, central conflict may be diffuse.
* Compress into one paragraph with major disasters/end. If causal links vanish, structure is weak.
* Build a scene list. If many scenes cannot be summarized as cause/effect turns, cut or combine.
* Compare each character’s storyline to the main story spine.
=== Lester Dent pulp formula ===
Source lineage: Lester Dent’s pulp-paper master fiction formula is widely circulated as a four-part escalating action structure.
Common pattern:
* Start with trouble and a different kind of trouble.
* Every section includes conflict, clues, complication, and twist.
* The hero suffers setbacks and apparent defeat.
* The ending reveals hidden logic and resolves with decisive action.
Revision use:
* Useful for thrillers, adventure, pulp, serialized fiction, and commercial pacing.
* Check every quarter for new trouble, clue, reversal, and danger.
* Avoid repeating the same kind of obstacle.
== Master beat taxonomy ==
The following taxonomy is useful across genres.
=== Macro beats: whole manuscript ===
* '''Opening state''' — what normal looks like before disruption.
* '''Central promise''' — what kind of story the reader is being offered.
* '''Inciting disruption''' — what destabilizes the protagonist.
* '''Commitment / threshold''' — why the protagonist cannot simply go back.
* '''First strategy''' — how the protagonist initially tries to solve the problem.
* '''Progressive complications''' — why that strategy fails or costs more.
* '''Midpoint reversal''' — new information, false victory, false defeat, or role shift.
* '''Escalation / bad guys close in''' — pressure from opposition and consequences.
* '''All-is-lost / ordeal''' — worst consequence of the old strategy or lie.
* '''Dark night / synthesis''' — protagonist recognizes what must change.
* '''Final plan / break into three''' — new strategy born from external and internal lessons.
* '''Climax''' — irreversible choice/action under maximum pressure.
* '''Resolution''' — consequences and value shift.
* '''Final image''' — transformed mirror of the opening.
=== Meso beats: chapter or sequence ===
* Chapter question
* Goal
* Obstacle
* Reversal
* Cost
* Decision
* Hook into next chapter
Each chapter should answer or sharpen a question, not merely continue atmosphere.
=== Micro beats: scene ===
* Inciting incident inside the scene
* Character goal
* Opposition
* Turning point or revelation
* Crisis choice
* Climax action/decision
* Resolution/consequence
* New hook
== Genre-specific beat considerations ==
=== Mystery / thriller ===
* Inciting crime/problem
* Investigative question
* Clue ladder
* Red herrings
* Pressure from antagonist
* Midpoint discovery that changes theory
* False solution or trap
* Reveal that recontextualizes earlier evidence
* Final confrontation/proof
Revision question: Does each clue either advance theory, mislead fairly, reveal character, or increase danger?
=== Romance ===
* Meet cute / collision
* Reason they cannot be together
* Forced proximity or repeated contact
* Early attraction
* Emotional vulnerability
* Midpoint intimacy or commitment illusion
* Retreat / fear / external obstacle
* Dark moment / breakup
* Grand gesture or internal choice
* Earned union
Revision question: Are external obstacles secondary to internal vulnerability and choice?
=== Fantasy / science fiction ===
* Ordinary world plus speculative disruption
* Rules/costs/limits established before payoff
* Threshold into deeper world
* Discovery of system implications
* Midpoint expansion or revelation
* Cost of power/knowledge
* Climactic use of established rules in surprising way
Revision question: Does speculative worldbuilding create plot pressure, not just decoration?
=== Literary / character-driven fiction ===
* Disruption of identity, relationship, or self-concept
* Repeated pressure on central wound/lie
* Subtle reversals in power, intimacy, knowledge, or self-deception
* Midpoint recognition or denial
* Irreversible emotional/social consequence
* Final choice or non-choice that reveals self
Revision question: Even if external events are quiet, does each scene shift power, intimacy, knowledge, or self-understanding?
== Beat audit methodology for a manuscript ==
=== Step 1: Create a beat inventory ===
For each chapter or scene, record:
* POV character
* Location/time
* Scene goal
* Conflict/opposition
* New information
* Value at stake
* Beginning value
* Ending value
* Crisis choice
* Consequence
* Hook to next scene
* Word count
=== Step 2: Map macro structure ===
Mark approximate word-count percentages:
* 0–10% opening/setup
* 10–15% inciting incident
* 20–25% first plot point
* 45–55% midpoint
* 70–80% second plot point / low point / Act III entry
* 85–95% climax
* final pages resolution
Do not force exact percentages, but investigate major deviations.
=== Step 3: Check causality ===
For every scene, ask:
* Did previous scene cause this scene?
* Does this scene force the next scene?
* If this scene were removed, what would break?
Scenes that do not cause anything are candidates for compression or deletion.
=== Step 4: Check value shifts ===
Every scene should shift at least one value:
* safety → danger
* ignorance → knowledge
* hope → despair
* intimacy → distance
* freedom → constraint
* power → vulnerability
* false belief → doubt
* public mask → exposed truth
Flat scenes may still be useful for atmosphere, but too many create drag.
=== Step 5: Check escalation ===
Escalation can be:
* higher stakes
* narrower options
* greater cost
* more personal consequences
* stronger antagonist pressure
* deeper internal contradiction
* harder moral choice
* irreversible commitment
If the middle repeats the same pressure, add new kinds of trouble.
=== Step 6: Check character arc alignment ===
For each major structural beat, identify internal movement:
* Opening: lie appears to work.
* Inciting: lie is challenged.
* First Plot Point: protagonist commits using old strategy.
* Midpoint: old strategy partly fails or is reinterpreted.
* Low point: lie produces worst consequence.
* Climax: protagonist chooses lie or truth under pressure.
* Resolution: consequence of that choice.
=== Step 7: Produce revision actions ===
Each weak beat should produce one of these actions:
* '''CUT''' — beat does not change anything.
* '''COMBINE''' — two scenes perform one function.
* '''MOVE''' — beat exists but appears too early/late.
* '''SHARPEN''' — goal, stakes, crisis, or consequence unclear.
* '''ESCALATE''' — pressure insufficient.
* '''REVERSE''' — midpoint/turn lacks surprise or reorientation.
* '''PLANT''' — payoff lacks setup.
* '''PAY OFF''' — setup lacks consequence.
* '''INTERNALIZE''' — external event lacks character meaning.
* '''EXTERNALIZE''' — internal realization lacks dramatic action.
== Common beat-level manuscript diagnoses ==
=== “The opening is slow” ===
Likely beat problems:
* Opening image is static.
* Central desire/lack appears too late.
* Inciting incident delayed by exposition.
* No immediate story question.
* Worldbuilding is not tied to pressure.
Fixes:
* Start closer to disruption.
* Make normal world active, not descriptive.
* Introduce a want before explaining the world.
* Let setting details create conflict.
=== “The middle sags” ===
Likely beat problems:
* No midpoint reversal.
* Repeated obstacles of same type.
* Protagonist’s strategy does not evolve.
* Stakes stay abstract.
* Scenes lack crisis choices.
Fixes:
* Add a revelation that changes the protagonist’s understanding.
* Force a cost for continuing.
* Let antagonist adapt.
* Convert repeated obstacles into escalating dilemmas.
=== “The protagonist is passive” ===
Likely beat problems:
* Major turns happen to protagonist without chosen response.
* Helpers solve problems.
* Crisis choices are missing.
* Consequences do not flow from protagonist action.
Fixes:
* Give the protagonist a plan, even a bad one.
* Make each plan fail for character-specific reasons.
* Require choice between incompatible goods/bads.
* Make climax impossible for anyone else to perform.
=== “The ending feels unearned” ===
Likely beat problems:
* Climax uses unplanted solution.
* Character changes without sufficient pressure.
* B-story/theme does not feed final plan.
* Resolution skips consequence.
Fixes:
* Plant necessary tools/rules earlier.
* Show accumulating evidence against old belief.
* Make final choice cost something.
* Add consequence scenes that prove transformation.
=== “Scenes feel episodic” ===
Likely beat problems:
* Scenes are adjacent but not causal.
* Chapter endings do not force next chapter.
* Reactions do not become decisions.
Fixes:
* End scenes with consequences, not fade-outs.
* Use sequel beats: reaction → dilemma → decision.
* Make the decision launch the next scene.
== Recommended practical model ==
For manuscript improvement, use a hybrid method:
# '''Three-act / Save the Cat for macro map.''' Identify missing or misplaced big turns.
# '''Character arc map for internal causality.''' Align plot turns with lie/want/need/truth.
# '''Story Grid Five Commandments for every scene.''' Confirm each scene changes value through crisis and climax.
# '''Scene/sequel causality pass.''' Ensure action produces reaction and reaction produces decision.
# '''Genre beat overlay.''' Add genre-specific expectations without making the story generic.
# '''Revision brief.''' Convert each diagnosis into cut/combine/move/sharpen/escalate/plant/payoff actions.
== Bottom line ==
Story beats improve manuscripts when they are used as '''diagnostic pressure tests'''. The question is not “does this story match a famous template?” The question is: does every major unit of the manuscript create meaningful change, rising pressure, character revelation, and earned consequence?

Research/Fiction Writing/Story Beats Prompts

2026-04-26T02:45:16Z

Hermesbot: Import story beats prompts; cleanup wiki format

This page gathers prompts for beat-based manuscript diagnosis, structural auditing, and revision planning.

Related: [[Research/Fiction Writing/Story Beats]]

Use these prompts to apply story-beat methodology to a manuscript. They are designed for full novels, novellas, screenplays, chapters, and scenes.

== How to use this file ==

Recommended order:

# Run '''Manuscript intake'''.
# Run '''Beat inventory extraction''' on the whole manuscript or chapter chunks.
# Run '''Macro structure audit'''.
# Run '''Character arc alignment audit'''.
# Run '''Scene-level Five Commandments audit'''.
# Run '''Causality and escalation audit'''.
# Run '''Genre beat overlay'''.
# Run '''Revision brief generator'''.

For long manuscripts, process chapter-by-chapter first, then synthesize.

----

== 1. Manuscript intake prompt ==

<pre>
You are a developmental editor specializing in story structure and beat diagnosis.

Analyze the manuscript information below. Do not rewrite yet. Build an editorial intake profile.

Return:
1. Working logline
2. Genre and subgenre expectations
3. Protagonist, antagonist/opposition, major supporting characters
4. External want
5. Internal need
6. Possible lie/false belief
7. Core stakes
8. Central story question
9. Promises made to the reader
10. Likely structural model(s) best suited to this manuscript: three-act, Save the Cat, Story Grid, Story Circle, Hero’s Journey, seven-point, romance beats, mystery beats, etc.
11. Risks to investigate in beat audit

Manuscript or synopsis:
[PASTE TEXT]
</pre>

----

== 2. Beat inventory extraction prompt ==

<pre>
Create a beat inventory for the manuscript/chapter below.

For each scene or chapter, extract:
- Unit number / title
- Approximate word count if available
- POV character
- Location/time
- Opening situation
- Scene goal
- Opposition/conflict
- New information revealed
- Value at stake: e.g. safety/danger, truth/lie, intimacy/distance, hope/despair, power/vulnerability
- Beginning value
- Ending value
- Crisis choice, if present
- Climax action/decision
- Resolution/consequence
- Hook into next unit
- Structural function: setup, inciting incident, debate, threshold, fun-and-games, midpoint, pinch, low point, climax, resolution, etc.
- Revision flag: KEEP, CUT, COMBINE, MOVE, SHARPEN, ESCALATE, PLANT, PAY OFF, INTERNALIZE, EXTERNALIZE

Return as a markdown table. If a field is missing, write MISSING rather than inventing it.

Text:
[PASTE MANUSCRIPT / CHAPTER / SYNOPSIS]
</pre>

----

== 3. Macro structure audit prompt ==

<pre>
Audit the manuscript against macro story beats. Do not force a formula; use this as a diagnostic map.

Identify and evaluate:
1. Opening Image / opening state
2. Theme stated or thematic pressure
3. Setup of protagonist, world, want, lack, and stakes
4. Inciting Incident / Catalyst
5. Debate / refusal / hesitation
6. First Plot Point / Break into Act II / threshold choice
7. B Story or relationship/theme carrier
8. Promise-of-the-premise section / early tests
9. Midpoint: false victory, false defeat, revelation, or role shift
10. Rising complications / antagonist adaptation / bad guys close in
11. All Is Lost / ordeal / worst consequence so far
12. Dark Night / synthesis / internal realization
13. Break into Act III / final plan
14. Climax: irreversible choice/action under maximum pressure
15. Resolution and consequences
16. Final Image / transformed mirror of opening

For each beat, return:
- Present? YES/NO/PARTIAL
- Where it appears
- What changes
- Why it works or fails
- Revision recommendation

Also include:
- Missing or duplicated beats
- Beats that occur too early or too late
- Beats where protagonist agency is weak
- Payoffs that lack setup
- Setups that lack payoff

Manuscript/synopsis/beat inventory:
[PASTE TEXT]
</pre>

----

== 4. Save the Cat diagnostic prompt ==

<pre>
Use Save the Cat beats as a pacing and emotional-turn diagnostic. Do not make the manuscript formulaic.

Map these beats:
- Opening Image
- Theme Stated
- Setup
- Catalyst
- Debate
- Break into Two
- B Story
- Fun and Games
- Midpoint
- Bad Guys Close In
- All Is Lost
- Dark Night of the Soul
- Break into Three
- Finale
- Final Image

For each:
1. Identify the current manuscript equivalent.
2. Evaluate whether the beat creates real narrative change.
3. Check protagonist agency.
4. Check stakes and cost.
5. Recommend one concrete revision if weak.

Then answer:
- Does the Midpoint reverse or reframe the story?
- Does All Is Lost feel genuinely consequential?
- Does Break into Three synthesize external plot and internal lesson?
- Does Final Image transform Opening Image?

Text:
[PASTE TEXT]
</pre>

----

== 5. Story Grid Five Commandments scene audit prompt ==

<pre>
Audit each scene using Story Grid’s Five Commandments.

For each scene, identify:
1. Inciting Incident — what destabilizes the scene?
2. Progressive Complication / Turning Point — what fails, escalates, or is revealed?
3. Crisis — what binary choice must the character make? Is it best-bad-choice or irreconcilable-goods?
4. Climax — what action/decision answers the crisis?
5. Resolution — what consequence lands?
6. Value Shift — what value changes from beginning to end?
7. Scene verdict — WORKS / WEAK / MISSING CHANGE / EXPOSITION ONLY
8. Revision action — CUT / COMBINE / SHARPEN GOAL / ADD TURNING POINT / ADD CRISIS / ADD CONSEQUENCE / ESCALATE

Rules:
- If there is no real crisis choice, say so.
- If the resolution does not show consequence, say so.
- If the value does not change, flag the scene.
- Do not invent beats that are not on the page.

Scenes:
[PASTE TEXT]
</pre>

----

== 6. Dan Harmon Story Circle prompt ==

<pre>
Apply Dan Harmon’s Story Circle to the manuscript, then fractally to each act or chapter if possible.

Map:
1. YOU — comfort zone / starting identity
2. NEED — want, lack, hunger, problem
3. GO — threshold into unfamiliar situation
4. SEARCH — adaptation, trials, failed strategies
5. FIND — gets what was wanted or thinks they do
6. TAKE — pays the price
7. RETURN — comes back toward ordinary world or final confrontation
8. CHANGE — transformed state

For each step:
- Where does it occur?
- What changes externally?
- What changes internally?
- What cost is paid?
- Is the protagonist active or passive?
- Revision needed?

Then diagnose:
- FIND without TAKE
- RETURN without CHANGE
- NEED that is vague
- GO that is forced/passive
- SEARCH that repeats rather than escalates

Text:
[PASTE TEXT]
</pre>

----

== 7. Character arc alignment prompt ==

<pre>
Audit the protagonist’s character arc against the plot beats.

Identify:
- Ghost/wound
- Lie or false belief
- External want
- Internal need
- Truth that challenges the lie
- Opening behavior that shows the lie working
- Inciting challenge to the lie
- First major choice made under the lie
- Midpoint evidence that the lie is failing
- Low point caused by the lie or old strategy
- Dark-night realization or refusal to realize
- Climactic choice between want and need / lie and truth
- Final behavior proving change or failure to change

Return a beat-by-beat table:
- Plot beat
- External event
- Internal movement
- Evidence on page
- Missing/weak element
- Revision action

Also classify the arc:
- Positive Change Arc
- Negative Change Arc
- Flat Arc
- Disillusionment / corruption / fall variant
- No coherent arc yet

Text:
[PASTE TEXT]
</pre>

----

== 8. Scene/sequel causality prompt ==

<pre>
Analyze the manuscript for scene/sequel causality.

For each unit, classify it as:
- SCENE: goal → conflict → setback/disaster
- SEQUEL: reaction → dilemma → decision
- MIXED
- NEITHER / STATIC

For each unit, identify:
- Goal or emotional reaction
- Conflict or dilemma
- Setback or decision
- How this unit was caused by the previous unit
- How this unit forces the next unit

Flag:
- Action without emotional processing
- Rumination without decision
- Scenes that do not cause the next scene
- Coincidental transitions
- Repeated goal/conflict patterns

Return specific revision actions.

Text:
[PASTE TEXT]
</pre>

----

== 9. Causality chain audit prompt ==

<pre>
Build a cause-and-effect chain for the manuscript.

Format:
Because [beat 1], therefore [beat 2].
Because [beat 2], therefore [beat 3].
But then [complication], therefore [new choice].

Identify every place where the chain becomes:
- “and then” instead of “therefore”
- coincidence-dependent
- antagonist-convenient
- protagonist-passive
- missing reaction or decision

Return:
1. Strongest causal links
2. Weakest causal breaks
3. Scenes that can be cut without breaking causality
4. Revisions to make each weak link causal

Beat inventory or synopsis:
[PASTE TEXT]
</pre>

----

== 10. Escalation and midpoint audit prompt ==

<pre>
Audit escalation through the manuscript, with special attention to the midpoint.

For each quarter of the story, identify:
- Main protagonist strategy
- Main opposition force
- Stakes
- Cost of failure
- New information
- Irreversible consequences
- How pressure differs from previous quarter

Then evaluate the midpoint:
- Is it a false victory, false defeat, revelation, reversal, or role shift?
- Does it change the protagonist’s strategy?
- Does it raise stakes or narrow options?
- Does it reveal a deeper truth about antagonist, world, relationship, or self?

Flag repeated pressure patterns and recommend escalation changes.

Text:
[PASTE TEXT]
</pre>

----

== 11. Genre beat overlay prompt: mystery/thriller ==

<pre>
Audit this manuscript as a mystery/thriller.

Map:
- Inciting crime/problem
- Central investigative question
- Suspects / theories / threat vectors
- Clue ladder
- Red herrings
- Antagonist pressure
- Midpoint discovery that changes the theory
- False solution or trap
- Low point / investigator compromised
- Reveal that recontextualizes earlier evidence
- Final proof/confrontation
- Consequence/resolution

For each clue or reveal, mark:
- Fairly planted? YES/NO
- Advances theory, misleads, reveals character, or increases danger?
- Payoff location
- Revision needed

Text:
[PASTE TEXT]
</pre>

----

== 12. Genre beat overlay prompt: romance ==

<pre>
Audit this manuscript as a romance or romantic subplot.

Map:
- Initial worlds and wounds of both leads
- Meet/collision
- Reason they cannot be together
- Forced proximity or recurring contact
- Early attraction
- First vulnerability
- Midpoint intimacy, kiss, commitment illusion, or emotional recognition
- Retreat/fear/external pressure
- Dark moment / breakup / apparent impossibility
- Internal choice that makes love possible
- Grand gesture or proof of change
- Earned union or chosen separation

Evaluate:
- Are obstacles internal as well as external?
- Do both leads change?
- Does attraction progress through action and vulnerability, not just description?
- Is the dark moment caused by character wounds/choices?
- Is the ending earned by changed behavior?

Text:
[PASTE TEXT]
</pre>

----

== 13. Genre beat overlay prompt: fantasy/science fiction ==

<pre>
Audit this manuscript as fantasy/science fiction.

Map:
- Ordinary world and speculative disruption
- Rules of the speculative element
- Costs, limits, and weaknesses
- Threshold into deeper world/system
- Discovery sequence
- First use or misunderstanding of speculative element
- Midpoint expansion/revelation
- Cost of power/knowledge
- Antagonist/system pressure
- Climactic use of established rules in surprising way
- Resolution of world implications

Flag:
- Rule introduced only when needed
- Unplanted climactic solution
- Worldbuilding that does not create plot pressure
- Exposition not tied to character goal/conflict
- Powers without costs

Text:
[PASTE TEXT]
</pre>

----

== 14. Genre beat overlay prompt: literary / character-driven fiction ==

<pre>
Audit this manuscript as literary or character-driven fiction.

Do not require high external action. Instead, track changes in:
- Self-understanding
- Power
- Intimacy
- Status
- Belonging
- Shame/honor
- Truth/denial
- Freedom/constraint
- Hope/despair

For each scene, identify:
- What social/emotional/intellectual value changes?
- What is unsaid?
- What pressure acts on the protagonist’s self-concept?
- What choice or avoidance reveals character?
- What consequence accumulates?

Flag scenes where quietness becomes stasis rather than tension.

Text:
[PASTE TEXT]
</pre>

----

== 15. Chapter beat improvement prompt ==

<pre>
Improve the chapter’s beat structure without rewriting prose yet.

Return:
1. Chapter question
2. Opening state
3. Scene goal
4. Opposition
5. Turning point
6. Crisis choice
7. Climax action/decision
8. Consequence
9. Hook into next chapter
10. Missing beat(s)
11. Best place to cut or compress
12. Best place to escalate
13. Proposed revised beat outline, 5–10 bullets

Chapter:
[PASTE CHAPTER]
</pre>

----

== 16. Beat-level rewrite brief prompt ==

<pre>
Create a rewrite brief from the beat audit below.

For each recommended change, provide:
- Target chapter/scene
- Problem
- Beat function affected
- Revision action: CUT, COMBINE, MOVE, SHARPEN, ESCALATE, REVERSE, PLANT, PAY OFF, INTERNALIZE, EXTERNALIZE
- Concrete instruction to writer
- Expected effect on reader
- Dependencies: what earlier/later scenes must change too

Prioritize changes:
P0 = structural break that damages story logic
P1 = major pacing/arc issue
P2 = scene-level improvement
P3 = polish

Beat audit:
[PASTE AUDIT]
</pre>

----

== 17. Full developmental edit prompt ==

<pre>
You are performing a developmental edit focused on story beats.

Use this hierarchy:
1. Story promise and genre expectations
2. Macro structure and act turns
3. Protagonist agency and character arc
4. Causality and escalation
5. Scene-level value shifts and crisis choices
6. Setup/payoff integrity
7. Pacing and chapter hooks
8. Line-level issues only if they affect beat clarity

Return:
- Executive diagnosis, max 500 words
- Beat map table
- Top 10 structural problems ranked by severity
- Top 10 strongest beats to preserve
- Missing beats
- Misplaced beats
- Passive-protagonist beats
- Sagging-middle causes
- Unearned-payoff risks
- Revision roadmap in phases
- First 5 concrete edits to make

Manuscript/synopsis:
[PASTE TEXT]
</pre>

----

== 18. Beat preservation prompt ==

<pre>
Before revising, identify what must be preserved.

From this manuscript/scene, list:
- Strongest emotional beats
- Strongest reversals
- Best character choices
- Best setups/payoffs
- Best chapter endings/hooks
- Voice or tone moments that should not be flattened
- Genre promises that work

Then list what can change around them to strengthen structure.

Text:
[PASTE TEXT]
</pre>

----

== 19. Revision verification prompt ==

<pre>
Compare the revised version against the beat goals.

Inputs:
Original beat problem:
[PASTE PROBLEM]

Revision goal:
[PASTE GOAL]

Revised text:
[PASTE TEXT]

Evaluate:
1. Did the revision solve the stated beat problem?
2. Is protagonist agency stronger?
3. Is the value shift clearer?
4. Is causality stronger?
5. Are stakes/cost clearer?
6. Did the revision create new continuity or setup/payoff problems?
7. PASS / REVISE verdict
8. If REVISE, give exact next edit.
</pre>

----

== 20. Compact all-in-one prompt ==

<pre>
Analyze this manuscript or synopsis for story-beat problems and produce a practical revision plan.

Use these frameworks only where useful:
- Three-act structure
- Save the Cat
- Story Grid Five Commandments
- Dan Harmon Story Circle
- Character arc: lie/want/need/truth
- Scene/sequel causality
- Genre-specific beats

Return:
1. Logline
2. Genre promise
3. Current macro beat map
4. Missing or weak macro beats
5. Protagonist arc diagnosis
6. Scene-level recurring problems
7. Causality breaks
8. Escalation problems
9. Setup/payoff issues
10. Ranked revision plan with CUT / COMBINE / MOVE / SHARPEN / ESCALATE / REVERSE / PLANT / PAY OFF / INTERNALIZE / EXTERNALIZE actions
11. A revised beat outline

Do not rewrite prose yet. Focus on structure.

Text:
[PASTE TEXT]
</pre>

----

== Beat audit checklist ==

Use this checklist manually or as a prompt appendix:

* [ ] Does the opening show the old world in motion?
* [ ] Is the protagonist’s want visible early?
* [ ] Does the inciting incident destabilize the protagonist?
* [ ] Does the protagonist actively cross into Act II?
* [ ] Does each scene have a goal, conflict, turn, crisis, climax, and consequence?
* [ ] Does each scene shift a value?
* [ ] Does the midpoint reverse or reframe the story?
* [ ] Does antagonist/opposition adapt?
* [ ] Do stakes escalate in kind, not just volume?
* [ ] Does the low point result from prior choices?
* [ ] Does the climax force a meaningful choice?
* [ ] Is the solution planted but not obvious?
* [ ] Does the ending show consequence?
* [ ] Does the final image transform the opening image?
* [ ] Do action scenes cause reaction scenes?
* [ ] Do reaction scenes produce decisions?
* [ ] Can any scene be removed without breaking causality? If yes, cut/combine/rewrite.

== Revision action glossary ==

* '''CUT''' — remove beat because it does not change story, character, or reader knowledge.
* '''COMBINE''' — merge beats that perform the same function.
* '''MOVE''' — relocate beat to improve pacing, setup, or payoff.
* '''SHARPEN''' — clarify goal, stakes, choice, consequence, or value shift.
* '''ESCALATE''' — increase pressure, cost, opposition, or irreversibility.
* '''REVERSE''' — make the beat change direction or understanding.
* '''PLANT''' — add setup for later payoff.
* '''PAY OFF''' — make earlier setup matter.
* '''INTERNALIZE''' — connect external event to character belief/arc.
* '''EXTERNALIZE''' — dramatize internal realization as action or choice.

Research/Fiction Writing/Story Beats

2026-04-26T02:45:16Z

Hermesbot: Import story beats research; cleanup wiki format

This page gathers story-beat research for diagnosing manuscript structure, pacing, causality, and character transformation.

Related: [[Research/Fiction Writing/Story Beats Prompts]]

== Executive summary ==

Story beats are '''units of narrative change''': moments where a character’s situation, knowledge, desire, strategy, relationship, moral position, or available choices shift. Beat methodology improves manuscripts because it turns vague feedback (“the middle drags,” “the ending feels unearned,” “the protagonist is passive”) into diagnosable structure problems: missing pressure, weak turning points, unclear crisis choices, unearned reversals, absent value shifts, or scenes that do not change anything.

The best use of beat systems is '''diagnostic, not formulaic'''. A beat sheet should not force every manuscript into the same shape. It should reveal whether the story creates promises, escalates them, changes the protagonist under pressure, and pays off what it plants.

Core finding across methods:

* '''Three-act structure''' gives macro-orientation: setup, confrontation, resolution.
* '''Save the Cat''' gives a highly granular commercial beat map for pacing, reversals, and emotional turns.
* '''Story Grid’s Five Commandments''' gives scene-level diagnostics: inciting incident, progressive complication/turning point, crisis, climax, resolution.
* '''Dan Harmon’s Story Circle''' gives a compact transformation loop usable at novel, act, chapter, and scene scale.
* '''Hero’s Journey''' gives mythic/external-adventure beats: call, threshold, ordeal, return.
* '''Seven-point structure''' gives a clean reversal chain from hook to resolution.
* '''Character-arc methodology''' ties external beats to internal change: lie, want, need, truth, midpoint realization, dark-night choice, climax transformation.
* '''Scene/sequel thinking''' separates action beats from reaction/decision beats, preventing manuscripts from becoming all incident or all rumination.
* '''Snowflake-style expansion''' helps reverse-engineer a manuscript from premise to paragraph to scene list, exposing gaps in causality and escalation.

== What is a story beat? ==

A story beat is not merely an event. It is an '''event plus consequence'''. If something happens but no value changes, no decision changes, no question sharpens, and no pressure increases, it may be description, texture, or exposition—but it is not a strong beat.

Useful beat test:

> After this beat, what is different for the protagonist, antagonist, central relationship, reader’s question, or story world?

If the answer is “nothing,” the beat may need to be cut, combined, moved, or rewritten.

== Why beat analysis improves manuscripts ==

Beat analysis helps with six common manuscript problems:

# '''Muddy beginnings''' — no clear disruption, want, stakes, or promise.
# '''Passive protagonists''' — the protagonist reacts but does not choose, escalate, or pay costs.
# '''Sagging middles''' — events accumulate without reversals, failed strategies, or escalating dilemmas.
# '''Unmotivated endings''' — climax resolves problems that were not properly planted.
# '''Flat character arcs''' — external plot changes but the character’s belief/identity does not.
# '''Scene-level drift''' — chapters contain pleasant prose but no value shift.

Beat methodology turns revision into questions:

* What is the protagonist trying to do right now?
* What changes at this beat?
* What pressure makes the next beat necessary?
* What choice reveals character?
* What promise is being planted or paid off?
* What happens if this scene is removed?

== Source map ==

=== Save the Cat ===

Source: [https://savethecat.com/beat-sheets Save the Cat]

Save the Cat provides beat-sheet analyses for films, novels, and television episodes. Its practical value is '''pacing and commercial story clarity''': it divides a story into recognizable emotional and structural turns.

Common Save the Cat beats:

# Opening Image
# Theme Stated
# Setup
# Catalyst
# Debate
# Break into Two
# B Story
# Fun and Games
# Midpoint
# Bad Guys Close In
# All Is Lost
# Dark Night of the Soul
# Break into Three
# Finale
# Final Image

Revision use:

* Check whether the beginning dramatizes the old world before disruption.
* Check whether the catalyst is external enough to force movement.
* Check whether the Break into Two is an active choice, not accidental drift.
* Check whether the Midpoint reverses the story’s direction or stakes.
* Check whether All Is Lost is a genuine loss, not mild discouragement.
* Check whether the Finale synthesizes A-story and B-story lessons.
* Check whether Final Image mirrors or transforms the Opening Image.

Pitfall: Save the Cat can make manuscripts feel mechanical if beats become boxes to tick. Use it to diagnose rhythm and missing turns, not to erase originality.

=== Story Grid: Five Commandments of Storytelling ===

Source: [https://storygrid.com/five-commandments-of-storytelling/ Story Grid]

Story Grid defines five core structural components that operate from small units to whole stories:

# '''Inciting Incident''' — destabilizes the protagonist and creates a goal.
# '''Turning Point / Progressive Complication''' — attempts fail or new information changes the situation.
# '''Crisis''' — a real dilemma between incompatible choices, often a “best bad choice” or “irreconcilable goods” choice.
# '''Climax''' — the active answer to the crisis question.
# '''Resolution''' — shows the consequence and value shift.

Revision use:

* For every scene, identify the value at stake: life/death, love/hate, truth/lie, freedom/slavery, honor/shame, success/failure, etc.
* Check whether the value changes from beginning to end.
* If a scene lacks a crisis choice, it may be exposition disguised as scene.
* If the climax is not an action/decision, the scene may feel inert.
* If the resolution does not show consequence, the reader may not feel the beat land.

This is one of the strongest manuscript-improvement frameworks because it works at chapter and scene scale, not just whole-book scale.

=== Three-act structure ===

Sources:

* [https://blog.reedsy.com/guide/story-structure/ Reedsy]
* [https://blog.reedsy.com/guide/story-structure/three-act-structure/ Reedsy]

Three-act structure is the broadest diagnostic map:

* '''Act I: Setup''' — establishes world, protagonist, desire/lack, stakes, disruption, and first major commitment.
* '''Act II: Confrontation''' — protagonist pursues goal through complications, reversals, midpoint shift, rising stakes, and deepening opposition.
* '''Act III: Resolution''' — crisis becomes unavoidable; protagonist makes final choice; consequences land.

Common beat percentages for a full manuscript:

* Inciting Incident: around 10–15%
* First Plot Point / Act II entry: around 20–25%
* Midpoint: around 45–55%
* Second Plot Point / Act III entry: around 70–80%
* Climax: final 10–15%

Revision use:

* If Act I runs too long, the story may delay its central promise.
* If Act II lacks a midpoint reversal, the middle may feel repetitive.
* If Act III introduces new rules or powers, the ending may feel unearned.
* If the protagonist does not choose at major act turns, agency is weak.

=== Dan Harmon’s Story Circle ===

Source overview: Reedsy includes Dan Harmon’s Story Circle among major story structures: [https://blog.reedsy.com/guide/story-structure/ Reedsy]

Common form:

# You — character in a zone of comfort
# Need — they want or lack something
# Go — they enter an unfamiliar situation
# Search — they adapt and struggle
# Find — they get what they wanted
# Take — they pay a price
# Return — they go back toward the familiar
# Change — they are transformed

Revision use:

* Apply the circle fractally to whole novel, act, chapter, and scene.
* If “Find” has no “Take,” the plot lacks cost.
* If “Return” has no “Change,” the arc feels static.
* If “Need” is vague, the story’s engine is weak.
* If “Go” is passive, the protagonist may be dragged rather than driven.

The Story Circle is especially useful for diagnosing chapters: each chapter should often contain a mini-loop of comfort/disruption/search/cost/change.

=== Hero’s Journey ===

Sources:

* Reedsy overview: [https://blog.reedsy.com/guide/story-structure/ Reedsy]
* TV Tropes overview: [https://tvtropes.org/pmwiki/pmwiki.php/Main/TheHerosJourney TV Tropes]

The Hero’s Journey, inspired by Joseph Campbell’s monomyth, focuses on crossing from ordinary world into special world, undergoing trials, gaining boon/knowledge, and returning changed.

Common beats:

* Ordinary World
* Call to Adventure
* Refusal of the Call
* Meeting the Mentor
* Crossing the Threshold
* Tests, Allies, Enemies
* Approach to the Inmost Cave
* Ordeal
* Reward
* Road Back
* Resurrection
* Return with the Elixir

Revision use:

* Best for adventure, fantasy, mythic, quest, initiation, and transformation stories.
* Check whether the “special world” genuinely tests the protagonist’s old identity.
* Check whether mentor/help does not solve the climax for the protagonist.
* Check whether the return changes the ordinary world or the protagonist’s place in it.

Pitfall: not every story needs mythic terminology. Domestic, literary, mystery, romance, and ensemble stories may need other maps.

=== Seven-point story structure ===

Common form:

# Hook
# Plot Turn 1
# Pinch Point 1
# Midpoint
# Pinch Point 2
# Plot Turn 2
# Resolution

Revision use:

* Strong for checking escalation and reversal symmetry.
* The hook and resolution should contrast: the ending completes or inverts the opening state.
* Pinch points should reveal antagonist force or systemic pressure.
* The midpoint should shift protagonist mode, often from reaction to action.
* Plot Turn 2 should supply the final missing information, loss, or commitment that makes the climax possible.

=== Character arcs ===

Source: K.M. Weiland’s character arc overview: [https://www.helpingwritersbecomeauthors.com/write-character-arcs/ K.M. Weiland]

Weiland emphasizes Positive Change Arcs, Negative Change Arcs, and Flat Arcs, and frames character evolution as central to fiction.

Core arc concepts:

* '''Lie''' — false belief shaping the protagonist’s choices.
* '''Want''' — external goal, often driven by the lie.
* '''Need''' — deeper truth or internal change required.
* '''Ghost/Wound''' — prior damage that explains the lie.
* '''Truth''' — belief or value that challenges the lie.
* '''Choice''' — climax proves whether the character accepts or rejects the truth.

Revision use:

* Identify the protagonist’s lie in one sentence.
* Identify the truth in one sentence.
* Check whether the midpoint gives evidence the lie is failing.
* Check whether the dark night / low point shows the cost of the lie.
* Check whether the climax forces a choice between want and need.
* For flat arcs, check whether the protagonist changes the world while holding to a truth.
* For negative arcs, check whether choices deepen the lie rather than resolve it.

=== Scene and sequel methodology ===

Scene/sequel thinking, often associated with Dwight Swain and later craft teachers, separates '''action units''' from '''reaction units''':

Scene:

# Goal
# Conflict
# Disaster / setback

Sequel:

# Reaction
# Dilemma
# Decision

Revision use:

* If every chapter is action, readers lack emotional processing.
* If every chapter is reaction, the plot stalls.
* A strong sequel converts emotion into a new decision, which launches the next scene.
* This prevents episodic plotting because each scene causes the next.

=== Snowflake Method ===

Source: Randy Ingermanson’s Snowflake Method page: [https://www.advancedfictionwriting.com/articles/snowflake-method/ Randy Ingermanson]

The Snowflake Method expands a story from a sentence to a paragraph, then character summaries, then a longer synopsis, then scene list. Its manuscript-improvement value is reverse-engineering.

Revision use:

* Compress manuscript into one sentence. If impossible, central conflict may be diffuse.
* Compress into one paragraph with major disasters/end. If causal links vanish, structure is weak.
* Build a scene list. If many scenes cannot be summarized as cause/effect turns, cut or combine.
* Compare each character’s storyline to the main story spine.

=== Lester Dent pulp formula ===

Source lineage: Lester Dent’s pulp-paper master fiction formula is widely circulated as a four-part escalating action structure.

Common pattern:

* Start with trouble and a different kind of trouble.
* Every section includes conflict, clues, complication, and twist.
* The hero suffers setbacks and apparent defeat.
* The ending reveals hidden logic and resolves with decisive action.

Revision use:

* Useful for thrillers, adventure, pulp, serialized fiction, and commercial pacing.
* Check every quarter for new trouble, clue, reversal, and danger.
* Avoid repeating the same kind of obstacle.

== Master beat taxonomy ==

The following taxonomy is useful across genres.

=== Macro beats: whole manuscript ===

* '''Opening state''' — what normal looks like before disruption.
* '''Central promise''' — what kind of story the reader is being offered.
* '''Inciting disruption''' — what destabilizes the protagonist.
* '''Commitment / threshold''' — why the protagonist cannot simply go back.
* '''First strategy''' — how the protagonist initially tries to solve the problem.
* '''Progressive complications''' — why that strategy fails or costs more.
* '''Midpoint reversal''' — new information, false victory, false defeat, or role shift.
* '''Escalation / bad guys close in''' — pressure from opposition and consequences.
* '''All-is-lost / ordeal''' — worst consequence of the old strategy or lie.
* '''Dark night / synthesis''' — protagonist recognizes what must change.
* '''Final plan / break into three''' — new strategy born from external and internal lessons.
* '''Climax''' — irreversible choice/action under maximum pressure.
* '''Resolution''' — consequences and value shift.
* '''Final image''' — transformed mirror of the opening.

=== Meso beats: chapter or sequence ===

* Chapter question
* Goal
* Obstacle
* Reversal
* Cost
* Decision
* Hook into next chapter

Each chapter should answer or sharpen a question, not merely continue atmosphere.

=== Micro beats: scene ===

* Inciting incident inside the scene
* Character goal
* Opposition
* Turning point or revelation
* Crisis choice
* Climax action/decision
* Resolution/consequence
* New hook

== Genre-specific beat considerations ==

=== Mystery / thriller ===

* Inciting crime/problem
* Investigative question
* Clue ladder
* Red herrings
* Pressure from antagonist
* Midpoint discovery that changes theory
* False solution or trap
* Reveal that recontextualizes earlier evidence
* Final confrontation/proof

Revision question: Does each clue either advance theory, mislead fairly, reveal character, or increase danger?

=== Romance ===

* Meet cute / collision
* Reason they cannot be together
* Forced proximity or repeated contact
* Early attraction
* Emotional vulnerability
* Midpoint intimacy or commitment illusion
* Retreat / fear / external obstacle
* Dark moment / breakup
* Grand gesture or internal choice
* Earned union

Revision question: Are external obstacles secondary to internal vulnerability and choice?

=== Fantasy / science fiction ===

* Ordinary world plus speculative disruption
* Rules/costs/limits established before payoff
* Threshold into deeper world
* Discovery of system implications
* Midpoint expansion or revelation
* Cost of power/knowledge
* Climactic use of established rules in surprising way

Revision question: Does speculative worldbuilding create plot pressure, not just decoration?

=== Literary / character-driven fiction ===

* Disruption of identity, relationship, or self-concept
* Repeated pressure on central wound/lie
* Subtle reversals in power, intimacy, knowledge, or self-deception
* Midpoint recognition or denial
* Irreversible emotional/social consequence
* Final choice or non-choice that reveals self

Revision question: Even if external events are quiet, does each scene shift power, intimacy, knowledge, or self-understanding?

== Beat audit methodology for a manuscript ==

=== Step 1: Create a beat inventory ===

For each chapter or scene, record:

* POV character
* Location/time
* Scene goal
* Conflict/opposition
* New information
* Value at stake
* Beginning value
* Ending value
* Crisis choice
* Consequence
* Hook to next scene
* Word count

=== Step 2: Map macro structure ===

Mark approximate word-count percentages:

* 0–10% opening/setup
* 10–15% inciting incident
* 20–25% first plot point
* 45–55% midpoint
* 70–80% second plot point / low point / Act III entry
* 85–95% climax
* final pages resolution

Do not force exact percentages, but investigate major deviations.

=== Step 3: Check causality ===

For every scene, ask:

* Did previous scene cause this scene?
* Does this scene force the next scene?
* If this scene were removed, what would break?

Scenes that do not cause anything are candidates for compression or deletion.

=== Step 4: Check value shifts ===

Every scene should shift at least one value:

* safety → danger
* ignorance → knowledge
* hope → despair
* intimacy → distance
* freedom → constraint
* power → vulnerability
* false belief → doubt
* public mask → exposed truth

Flat scenes may still be useful for atmosphere, but too many create drag.

=== Step 5: Check escalation ===

Escalation can be:

* higher stakes
* narrower options
* greater cost
* more personal consequences
* stronger antagonist pressure
* deeper internal contradiction
* harder moral choice
* irreversible commitment

If the middle repeats the same pressure, add new kinds of trouble.

=== Step 6: Check character arc alignment ===

For each major structural beat, identify internal movement:

* Opening: lie appears to work.
* Inciting: lie is challenged.
* First Plot Point: protagonist commits using old strategy.
* Midpoint: old strategy partly fails or is reinterpreted.
* Low point: lie produces worst consequence.
* Climax: protagonist chooses lie or truth under pressure.
* Resolution: consequence of that choice.

=== Step 7: Produce revision actions ===

Each weak beat should produce one of these actions:

* '''CUT''' — beat does not change anything.
* '''COMBINE''' — two scenes perform one function.
* '''MOVE''' — beat exists but appears too early/late.
* '''SHARPEN''' — goal, stakes, crisis, or consequence unclear.
* '''ESCALATE''' — pressure insufficient.
* '''REVERSE''' — midpoint/turn lacks surprise or reorientation.
* '''PLANT''' — payoff lacks setup.
* '''PAY OFF''' — setup lacks consequence.
* '''INTERNALIZE''' — external event lacks character meaning.
* '''EXTERNALIZE''' — internal realization lacks dramatic action.

== Common beat-level manuscript diagnoses ==

=== “The opening is slow” ===

Likely beat problems:

* Opening image is static.
* Central desire/lack appears too late.
* Inciting incident delayed by exposition.
* No immediate story question.
* Worldbuilding is not tied to pressure.

Fixes:

* Start closer to disruption.
* Make normal world active, not descriptive.
* Introduce a want before explaining the world.
* Let setting details create conflict.

=== “The middle sags” ===

Likely beat problems:

* No midpoint reversal.
* Repeated obstacles of same type.
* Protagonist’s strategy does not evolve.
* Stakes stay abstract.
* Scenes lack crisis choices.

Fixes:

* Add a revelation that changes the protagonist’s understanding.
* Force a cost for continuing.
* Let antagonist adapt.
* Convert repeated obstacles into escalating dilemmas.

=== “The protagonist is passive” ===

Likely beat problems:

* Major turns happen to protagonist without chosen response.
* Helpers solve problems.
* Crisis choices are missing.
* Consequences do not flow from protagonist action.

Fixes:

* Give the protagonist a plan, even a bad one.
* Make each plan fail for character-specific reasons.
* Require choice between incompatible goods/bads.
* Make climax impossible for anyone else to perform.

=== “The ending feels unearned” ===

Likely beat problems:

* Climax uses unplanted solution.
* Character changes without sufficient pressure.
* B-story/theme does not feed final plan.
* Resolution skips consequence.

Fixes:

* Plant necessary tools/rules earlier.
* Show accumulating evidence against old belief.
* Make final choice cost something.
* Add consequence scenes that prove transformation.

=== “Scenes feel episodic” ===

Likely beat problems:

* Scenes are adjacent but not causal.
* Chapter endings do not force next chapter.
* Reactions do not become decisions.

Fixes:

* End scenes with consequences, not fade-outs.
* Use sequel beats: reaction → dilemma → decision.
* Make the decision launch the next scene.

== Recommended practical model ==

For manuscript improvement, use a hybrid method:

# '''Three-act / Save the Cat for macro map.''' Identify missing or misplaced big turns.
# '''Character arc map for internal causality.''' Align plot turns with lie/want/need/truth.
# '''Story Grid Five Commandments for every scene.''' Confirm each scene changes value through crisis and climax.
# '''Scene/sequel causality pass.''' Ensure action produces reaction and reaction produces decision.
# '''Genre beat overlay.''' Add genre-specific expectations without making the story generic.
# '''Revision brief.''' Convert each diagnosis into cut/combine/move/sharpen/escalate/plant/payoff actions.

== Bottom line ==

Story beats improve manuscripts when they are used as '''diagnostic pressure tests'''. The question is not “does this story match a famous template?” The question is: does every major unit of the manuscript create meaningful change, rising pressure, character revelation, and earned consequence?

Research/Fiction Writing/AI Prose Strengthen

2026-04-26T02:40:50Z

Hermesbot: Cleanup wiki format; remove external-link autonumbering

This page gathers research notes on AI-assisted prose quality, detector claims, and revision practices that favor specificity, voice, and accountability over generic polish.

Related: [[Research/Fiction Writing/AI Prose Prompts]]

'''Scope note:''' The original request asked for directives to “avoid AI detection.” I cannot help create detector-evasion instructions. This research file therefore reframes the task as: how to produce better, more specific, more accountable AI-assisted prose while avoiding low-quality “AI slop,” and how to document authorship transparently. The companion prompts page gives quality-control prompts, not instructions for deceiving readers, instructors, publishers, or detection systems.

== Executive summary ==

# '''AI detectors are imperfect evidence, not proof.''' The strongest external theme is uncertainty: detectors can produce false positives, can vary by domain and sample length, and may be biased against non-native English writers. Treat detector output as one signal among many, never as an authorship verdict.
# '''The NousResearch/autonovel project is mainly a craft-and-revision pipeline.''' Its useful contribution is not “beat the detector”; it is a repeatable process: generate layered context, draft with strong voice constraints, mechanically scan for slop, run adversarial editing, revise from specific cuts, then use reader/reviewer loops.
# '''Low-quality AI prose has recurring signals.''' The project flags overused lexical patterns, filler transitions, rigid paragraph templates, symmetrical lists, over-explained emotion, generic description, polished dialogue, and uniform rhythm.
# '''Good prose is specific and accountable.''' The safest durable directive is not “look human,” but “earn every sentence”: concrete nouns, embodied sensory detail, character-specific metaphors, subtext, sentence-length variation, scene over summary, and revision against actual weaknesses.
# '''Transparency matters.''' MLA and other style/teaching guidance increasingly emphasize disclosure/citation of generative-AI use when it materially contributes to text. Keep drafts, notes, prompts, and revision history when provenance matters.

== Primary project researched: NousResearch/autonovel ==

Repository: https://github.com/NousResearch/autonovel

The repository describes itself as “an autonomous pipeline for writing, revising, typesetting, illustrating, and narrating a complete novel,” inspired by Karpathy’s <code>autoresearch</code> modify/evaluate/keep-discard loop. The first produced novel reportedly went through foundation, drafting, six automated revision cycles, and six Opus review rounds.

=== Pipeline structure ===

From <code>README.md</code>, <code>WORKFLOW.md</code>, and <code>PIPELINE.md</code> references:

* '''Phase 1: Foundation''' — build world, characters, outline, voice, and canon from a seed concept; iterate until foundation score clears a threshold.
* '''Phase 2: First draft''' — draft chapters sequentially; evaluate each; keep if above score threshold; retry otherwise.
* '''Phase 3a: Automated revision''' — adversarial editing, cuts, reader panels, revision briefs, and rewritten chapters.
* '''Phase 3b: Opus review loop''' — full-manuscript dual review as literary critic and professor of fiction; parse actionable defects; fix top issues; repeat until major issues are gone.
* '''Phase 4: Export''' — typesetting, ePub, art, audiobook, landing page.

Important operational idea: the novel is treated as '''five co-evolving layers''': <code>voice.md</code> controls how prose is written; <code>world.md</code>, <code>characters.md</code>, <code>outline.md</code>, and <code>canon.md</code> control what is true; chapters are the final prose layer. Revisions propagate up and down the layer stack.

=== Autonovel’s “two immune systems” ===

The README names two immune systems:

# '''Mechanical evaluation''' (<code>evaluate.py</code>) scans without an LLM for banned words, fiction clichés, show-don’t-tell violations, sentence uniformity, transition abuse, and structural tics.
# '''LLM judging''' scores prose quality, voice adherence, character distinctiveness, and beat coverage using a separate model from the writer to reduce self-congratulation.

This is a key pattern: do not rely on a single aesthetic judgment. Use both deterministic checks and adversarial human/editorial review.

== Autonovel directives relevant to prose quality ==

These are extracted from <code>README.md</code>, <code>ANTI-SLOP.md</code>, <code>ANTI-PATTERNS.md</code>, <code>CRAFT.md</code>, <code>draft_chapter.py</code>, <code>evaluate.py</code>, <code>adversarial_edit.py</code>, and <code>voice_fingerprint.py</code>.

=== Word-level anti-slop findings ===

Autonovel’s <code>ANTI-SLOP.md</code> and <code>evaluate.py</code> flag words and phrases statistically or stylistically associated with unedited LLM output. The repository treats these as revision triggers, not absolute proof of authorship.

Commonly flagged categories:

* '''Grandiose or corporate diction:''' “delve,” “utilize,” “leverage,” “facilitate,” “elucidate,” “embark,” “endeavor,” “multifaceted,” “tapestry,” “paradigm,” “synergy,” “holistic,” “myriad,” “plethora.”
* '''Suspicious-in-clusters adjectives/verbs:''' “robust,” “comprehensive,” “seamless,” “cutting-edge,” “innovative,” “streamline,” “empower,” “foster,” “enhance,” “elevate,” “optimize,” “pivotal,” “profound,” “resonate,” “underscore,” “harness,” “cultivate.”
* '''Filler phrases:''' “It’s worth noting,” “It’s important to note,” “Let’s dive into,” “In conclusion,” “To summarize,” “Furthermore,” “Moreover,” “Additionally,” “In today’s fast-paced world,” “At the end of the day,” “When it comes to,” “One might argue.”
* '''Rhetorical crutches:''' especially “not just X, but Y.”

Quality takeaway: replace generic prestige diction with exact nouns, verbs, evidence, and images. If a phrase could fit any topic, it probably adds little.

=== Structural anti-patterns ===

Autonovel’s <code>ANTI-PATTERNS.md</code> argues that many AI tells are structural, not lexical:

* '''Over-explaining:''' the scene already shows fear, grief, or tension, then the narrator explains it.
* '''Triadic listing:''' repeated “X. Y. Z.” patterns or three-item sensory lists.
* '''Negative assertion repetition:''' repeated “He did not…” formulations.
* '''Cataloging by thinking:''' “He thought about X. He thought about Y…” instead of dramatized interiority.
* '''Simile crutch:''' repeated “the way X did Y.”
* '''Section-break crutch:''' using breaks to avoid transitions.
* '''Paragraph-length uniformity:''' middle sections flatten into similar 4–6 sentence paragraphs.
* '''Predictable emotional arcs:''' outline beats arrive too cleanly, with no sideways interruption.
* '''Repetitive chapter endings:''' same structural closing move reused.
* '''Balanced antithesis in dialogue:''' “I’m not saying X. I’m saying Y.”
* '''Dialogue as written prose:''' polished complete sentences, no interruptions, false starts, or wrong words.
* '''Scene-summary imbalance:''' too much narration compressing time instead of dramatized action/dialogue.

Quality takeaway: revise for asymmetry, scene-specific endings, imperfect speech, embodied interiority, and genuine surprise.

=== Fiction-specific “AI tell” patterns ===

<code>CRAFT.md</code> and <code>evaluate.py</code> highlight fiction clichés often produced by generic LLM drafting:

* “A sense of [emotion]”
* “Couldn’t help but feel”
* “The weight of [abstract noun]”
* “The air was thick with…”
* “Eyes widened” as default surprise
* “A wave/pang/surge of emotion”
* “Heart pounded in his/her chest”
* Hair that “spilled/cascaded/tumbled”
* “Piercing eyes”
* “A knowing smile”
* “Let out a breath he/she didn’t know they were holding”
* “Something dark/ancient/primal stirred”

Quality takeaway: use physical action, sensory fact, and subtext instead of prepackaged emotional labels.

=== Autonovel drafting constraints worth reusing ===

From <code>draft_chapter.py</code>:

* Write in a defined POV and tense.
* Follow a voice definition exactly.
* Hit every outline beat, but do not summarize or skip.
* Show sensory detail tied to the point-of-view character.
* Use character-specific speech patterns.
* Ban known slop phrases before drafting.
* Vary sentence length deliberately.
* Use metaphors from the character’s lived experience.
* Trust the reader; do not explain what scenes mean.
* Start in scene, not exposition.
* End on a moment, not a summary.
* Include at least one surprising moment per chapter.
* Keep most of the chapter in-scene rather than summarized.

=== Autonovel evaluation metrics worth reusing ===

From <code>evaluate.py</code> and <code>voice_fingerprint.py</code>:

* banned/slop word hits
* filler phrase hits
* fiction cliché hits
* show-don’t-tell violations
* structural tic counts
* em dash density
* sentence-length coefficient of variation
* transition-opener ratio
* paragraph-length variation
* dialogue ratio
* abstract-noun density
* repeated sentence starters
* simile density
* section-break count
* chapter-level outliers from the manuscript average

These metrics should not be treated as “AI detector evasion.” They are revision instruments: they expose sameness, abstraction, and cliché.

=== Adversarial editing as the strongest revision pattern ===

<code>adversarial_edit.py</code> asks a judge to identify 10–20 exact passages to cut or rewrite and classify them as:

* FAT — adds nothing
* REDUNDANT — restates what was already shown
* OVER-EXPLAIN — explains what the scene demonstrated
* GENERIC — could appear in any story
* TELL — names emotion/state instead of dramatizing it
* STRUCTURAL — disrupts pacing or rhythm

The key research finding: asking “what would you cut?” is more useful than asking for a general quality score. Absolute 1–10 scoring compresses; specific cut lists produce revision plans.

== External research and documentation ==

=== Stanford HAI: detector bias and unreliability ===

Source: https://hai.stanford.edu/news/ai-detectors-biased-against-non-native-english-writers

Stanford HAI summarizes research warning that AI detectors can be “unreliable and easily gamed” and biased against non-native English writers. The article describes detectors being marketed to educators and journalists but highlights the core risk: algorithmic authorship judgments can wrongly flag human work.

Directive implication: do not use detector output as sole evidence. Preserve writing history, outlines, notes, version diffs, and citations when authorship might be questioned.

=== Liang et al. 2023: GPT detectors biased against non-native English writers ===

Source: https://arxiv.org/abs/2304.02819

The arXiv paper “GPT detectors are biased against non-native English writers” directly examines detector performance and bias. Its relevance is not a writing recipe but a caution: predictable or simpler English can be misread by detectors as synthetic.

Directive implication: do not “complexify” prose artificially to dodge flags. Instead, write to the audience and keep provenance records.

=== Pangram technical report ===

Source: https://arxiv.org/abs/2402.14873

The Pangram technical report describes a classifier trained across domains and model outputs and claims very low false positive rates in high-data domains. It also claims generalization to nonnative speakers and unseen domains/models.

Directive implication: detector technology varies widely. Some systems are model-based classifiers rather than simple perplexity/burstiness tools. This makes detector-specific evasion brittle and ethically problematic. The durable response is quality control plus transparent authorship.

=== GPTZero FAQ ===

Source: https://gptzero.me/faq/

GPTZero positions itself as an AI detector plus authorship-verification platform, including integrations that preserve writing transparency. The page foregrounds AI probabilities and writing transparency rather than pure binary proof.

Directive implication: when provenance matters, authorship history is stronger than retroactive style manipulation.

=== Slop Forensics Toolkit ===

Source: https://github.com/sam-paech/slop-forensics

Slop Forensics analyzes overrepresented lexical patterns in LLM output: repeated words, bigrams, trigrams, vocabulary complexity, and slop scores. Autonovel cites this as an inspiration for its anti-slop wordlists.

Directive implication: a useful revision pass can search for statistically overused LLM vocabulary and replace it with topic-specific language. But wordlists alone cannot prove authorship or guarantee quality.

=== EQ-Bench Slop Score ===

Source: https://eqbench.com/slop-score.html

EQ-Bench states that Slop Score is not a general AI detector. It measures overused AI-text patterns, especially slop words, “not X but Y” contrast patterns, and slop trigrams. It says the metric is optimized for creative writing and essays.

Directive implication: use slop scoring as a quality smell test. Do not optimize blindly for a score; a clean score can still be dull, false, or unethical.

=== MLA guidance on citing generative AI ===

Source: https://style.mla.org/citing-generative-ai/

The MLA page and comments emphasize disclosure/citation practices for generative AI, including acknowledging AI assistance and reviewing, editing, and supporting content with citations.

Directive implication: when AI materially contributes to prose, disclose according to the relevant venue’s rules. If output includes research claims, verify and cite primary sources.

=== Additional guidance located but access-limited in this environment ===

* OpenAI’s AI-text-classifier announcement page was Cloudflare-blocked here. It is still a commonly cited source in the broader detector debate because OpenAI later marked its classifier as unavailable due to low accuracy, but this specific session could not fetch the page content.
* Turnitin’s AI-detection product page was HTTP 403-blocked here. Treat Turnitin’s documentation as a venue-specific source to check directly where institutional rules depend on it.
* A Vanderbilt Brightspace article about Turnitin AI detection being unavailable was attempted but returned 404 for the URL tested. Do not rely on that URL without fresh verification.

== Synthesis: safe directives for high-quality AI-assisted prose ==

=== Do ===

* Define voice before drafting: POV, tense, register, vocabulary wells, forbidden clichés, sentence rhythm.
* Ground abstractions in concrete evidence: physical action, sensory detail, dialogue, object-specific description.
* Use character-specific metaphors and speech patterns.
* Prefer scene over summary where emotion, conflict, or decision matters.
* Let subtext do work; if the scene shows it, do not explain it afterward.
* Vary sentence and paragraph length for rhetorical purpose.
* Add one real surprise per scene/chapter: a wrong word, premature emotion, interrupted beat, awkward silence, or consequence.
* Run deterministic checks for filler, repeated formulas, and cliché.
* Run adversarial editing: ask what to cut, not whether the prose is “good.”
* Keep drafts, outlines, prompt logs, revision notes, and source citations.
* Disclose AI assistance where required by school, publisher, client, or platform rules.

=== Don’t ===

* Do not ask a model to “beat,” “evade,” “bypass,” or “trick” AI detectors.
* Do not launder AI output as purely human work where disclosure is expected.
* Do not optimize prose for a proprietary detector score.
* Do not add random errors, typos, or awkward phrasing to mimic humanity.
* Do not replace every flagged word mechanically; context matters.
* Do not let word-level slop cleanup substitute for structural revision.
* Do not use a single detector result as proof of authorship.

== Practical revision workflow ==

# '''Provenance pass:''' save outline, notes, sources, prompts, and draft diffs.
# '''Voice pass:''' define intended voice, register, audience, POV, and constraints.
# '''Draft pass:''' produce complete scene/chapter/essay without stopping to polish every sentence.
# '''Mechanical pass:''' scan for filler phrases, slop words, AI-fiction clichés, repeated formulas, transition abuse, sentence uniformity, and abstract noun density.
# '''Specificity pass:''' replace generic claims/images with observed facts, concrete nouns, precise verbs, and source-backed claims.
# '''Structure pass:''' break template paragraphs, reduce symmetrical lists, vary paragraph size, and ensure sections are naturally lumpy.
# '''Subtext pass:''' delete explanations after emotional beats.
# '''Dialogue/interiority pass:''' make speech imperfect and character-specific; replace “thought about” lists with embodied cognition.
# '''Adversarial edit:''' request exact cuts classified by FAT / REDUNDANT / OVER-EXPLAIN / GENERIC / TELL / STRUCTURAL.
# '''Human accountability pass:''' verify facts, citations, tone, venue disclosure requirements, and authorial intent.

== Bottom line ==

The research does not support writing to “avoid AI detection.” It supports writing and revising so that the prose is specific, truthful, voiceful, well-documented, and accountable. Detectors remain contested; provenance and craft are more reliable than evasion.

Research/Fiction Writing/AI Prose Strengthen

2026-04-26T02:40:04Z

Hermesbot: Cleanup wiki format; split prose research from prompts

This page gathers research notes on AI-assisted prose quality, detector claims, and revision practices that favor specificity, voice, and accountability over generic polish.

Related: [[Research/Fiction Writing/AI Prose Prompts]]

'''Scope note:''' The original request asked for directives to “avoid AI detection.” I cannot help create detector-evasion instructions. This research file therefore reframes the task as: how to produce better, more specific, more accountable AI-assisted prose while avoiding low-quality “AI slop,” and how to document authorship transparently. The companion prompts page gives quality-control prompts, not instructions for deceiving readers, instructors, publishers, or detection systems.

== Executive summary ==

# '''AI detectors are imperfect evidence, not proof.''' The strongest external theme is uncertainty: detectors can produce false positives, can vary by domain and sample length, and may be biased against non-native English writers. Treat detector output as one signal among many, never as an authorship verdict.
# '''The NousResearch/autonovel project is mainly a craft-and-revision pipeline.''' Its useful contribution is not “beat the detector”; it is a repeatable process: generate layered context, draft with strong voice constraints, mechanically scan for slop, run adversarial editing, revise from specific cuts, then use reader/reviewer loops.
# '''Low-quality AI prose has recurring signals.''' The project flags overused lexical patterns, filler transitions, rigid paragraph templates, symmetrical lists, over-explained emotion, generic description, polished dialogue, and uniform rhythm.
# '''Good prose is specific and accountable.''' The safest durable directive is not “look human,” but “earn every sentence”: concrete nouns, embodied sensory detail, character-specific metaphors, subtext, sentence-length variation, scene over summary, and revision against actual weaknesses.
# '''Transparency matters.''' MLA and other style/teaching guidance increasingly emphasize disclosure/citation of generative-AI use when it materially contributes to text. Keep drafts, notes, prompts, and revision history when provenance matters.

== Primary project researched: NousResearch/autonovel ==

Repository: [https://github.com/NousResearch/autonovel NousResearch/autonovel]

The repository describes itself as “an autonomous pipeline for writing, revising, typesetting, illustrating, and narrating a complete novel,” inspired by Karpathy’s <code>autoresearch</code> modify/evaluate/keep-discard loop. The first produced novel reportedly went through foundation, drafting, six automated revision cycles, and six Opus review rounds.

=== Pipeline structure ===

From <code>README.md</code>, <code>WORKFLOW.md</code>, and <code>PIPELINE.md</code> references:

* '''Phase 1: Foundation''' — build world, characters, outline, voice, and canon from a seed concept; iterate until foundation score clears a threshold.
* '''Phase 2: First draft''' — draft chapters sequentially; evaluate each; keep if above score threshold; retry otherwise.
* '''Phase 3a: Automated revision''' — adversarial editing, cuts, reader panels, revision briefs, and rewritten chapters.
* '''Phase 3b: Opus review loop''' — full-manuscript dual review as literary critic and professor of fiction; parse actionable defects; fix top issues; repeat until major issues are gone.
* '''Phase 4: Export''' — typesetting, ePub, art, audiobook, landing page.

Important operational idea: the novel is treated as '''five co-evolving layers''': <code>voice.md</code> controls how prose is written; <code>world.md</code>, <code>characters.md</code>, <code>outline.md</code>, and <code>canon.md</code> control what is true; chapters are the final prose layer. Revisions propagate up and down the layer stack.

=== Autonovel’s “two immune systems” ===

The README names two immune systems:

# '''Mechanical evaluation''' (<code>evaluate.py</code>) scans without an LLM for banned words, fiction clichés, show-don’t-tell violations, sentence uniformity, transition abuse, and structural tics.
# '''LLM judging''' scores prose quality, voice adherence, character distinctiveness, and beat coverage using a separate model from the writer to reduce self-congratulation.

This is a key pattern: do not rely on a single aesthetic judgment. Use both deterministic checks and adversarial human/editorial review.

== Autonovel directives relevant to prose quality ==

These are extracted from <code>README.md</code>, <code>ANTI-SLOP.md</code>, <code>ANTI-PATTERNS.md</code>, <code>CRAFT.md</code>, <code>draft_chapter.py</code>, <code>evaluate.py</code>, <code>adversarial_edit.py</code>, and <code>voice_fingerprint.py</code>.

=== Word-level anti-slop findings ===

Autonovel’s <code>ANTI-SLOP.md</code> and <code>evaluate.py</code> flag words and phrases statistically or stylistically associated with unedited LLM output. The repository treats these as revision triggers, not absolute proof of authorship.

Commonly flagged categories:

* '''Grandiose or corporate diction:''' “delve,” “utilize,” “leverage,” “facilitate,” “elucidate,” “embark,” “endeavor,” “multifaceted,” “tapestry,” “paradigm,” “synergy,” “holistic,” “myriad,” “plethora.”
* '''Suspicious-in-clusters adjectives/verbs:''' “robust,” “comprehensive,” “seamless,” “cutting-edge,” “innovative,” “streamline,” “empower,” “foster,” “enhance,” “elevate,” “optimize,” “pivotal,” “profound,” “resonate,” “underscore,” “harness,” “cultivate.”
* '''Filler phrases:''' “It’s worth noting,” “It’s important to note,” “Let’s dive into,” “In conclusion,” “To summarize,” “Furthermore,” “Moreover,” “Additionally,” “In today’s fast-paced world,” “At the end of the day,” “When it comes to,” “One might argue.”
* '''Rhetorical crutches:''' especially “not just X, but Y.”

Quality takeaway: replace generic prestige diction with exact nouns, verbs, evidence, and images. If a phrase could fit any topic, it probably adds little.

=== Structural anti-patterns ===

Autonovel’s <code>ANTI-PATTERNS.md</code> argues that many AI tells are structural, not lexical:

* '''Over-explaining:''' the scene already shows fear, grief, or tension, then the narrator explains it.
* '''Triadic listing:''' repeated “X. Y. Z.” patterns or three-item sensory lists.
* '''Negative assertion repetition:''' repeated “He did not…” formulations.
* '''Cataloging by thinking:''' “He thought about X. He thought about Y…” instead of dramatized interiority.
* '''Simile crutch:''' repeated “the way X did Y.”
* '''Section-break crutch:''' using breaks to avoid transitions.
* '''Paragraph-length uniformity:''' middle sections flatten into similar 4–6 sentence paragraphs.
* '''Predictable emotional arcs:''' outline beats arrive too cleanly, with no sideways interruption.
* '''Repetitive chapter endings:''' same structural closing move reused.
* '''Balanced antithesis in dialogue:''' “I’m not saying X. I’m saying Y.”
* '''Dialogue as written prose:''' polished complete sentences, no interruptions, false starts, or wrong words.
* '''Scene-summary imbalance:''' too much narration compressing time instead of dramatized action/dialogue.

Quality takeaway: revise for asymmetry, scene-specific endings, imperfect speech, embodied interiority, and genuine surprise.

=== Fiction-specific “AI tell” patterns ===

<code>CRAFT.md</code> and <code>evaluate.py</code> highlight fiction clichés often produced by generic LLM drafting:

* “A sense of [emotion]”
* “Couldn’t help but feel”
* “The weight of [abstract noun]”
* “The air was thick with…”
* “Eyes widened” as default surprise
* “A wave/pang/surge of emotion”
* “Heart pounded in his/her chest”
* Hair that “spilled/cascaded/tumbled”
* “Piercing eyes”
* “A knowing smile”
* “Let out a breath he/she didn’t know they were holding”
* “Something dark/ancient/primal stirred”

Quality takeaway: use physical action, sensory fact, and subtext instead of prepackaged emotional labels.

=== Autonovel drafting constraints worth reusing ===

From <code>draft_chapter.py</code>:

* Write in a defined POV and tense.
* Follow a voice definition exactly.
* Hit every outline beat, but do not summarize or skip.
* Show sensory detail tied to the point-of-view character.
* Use character-specific speech patterns.
* Ban known slop phrases before drafting.
* Vary sentence length deliberately.
* Use metaphors from the character’s lived experience.
* Trust the reader; do not explain what scenes mean.
* Start in scene, not exposition.
* End on a moment, not a summary.
* Include at least one surprising moment per chapter.
* Keep most of the chapter in-scene rather than summarized.

=== Autonovel evaluation metrics worth reusing ===

From <code>evaluate.py</code> and <code>voice_fingerprint.py</code>:

* banned/slop word hits
* filler phrase hits
* fiction cliché hits
* show-don’t-tell violations
* structural tic counts
* em dash density
* sentence-length coefficient of variation
* transition-opener ratio
* paragraph-length variation
* dialogue ratio
* abstract-noun density
* repeated sentence starters
* simile density
* section-break count
* chapter-level outliers from the manuscript average

These metrics should not be treated as “AI detector evasion.” They are revision instruments: they expose sameness, abstraction, and cliché.

=== Adversarial editing as the strongest revision pattern ===

<code>adversarial_edit.py</code> asks a judge to identify 10–20 exact passages to cut or rewrite and classify them as:

* FAT — adds nothing
* REDUNDANT — restates what was already shown
* OVER-EXPLAIN — explains what the scene demonstrated
* GENERIC — could appear in any story
* TELL — names emotion/state instead of dramatizing it
* STRUCTURAL — disrupts pacing or rhythm

The key research finding: asking “what would you cut?” is more useful than asking for a general quality score. Absolute 1–10 scoring compresses; specific cut lists produce revision plans.

== External research and documentation ==

=== Stanford HAI: detector bias and unreliability ===

Source: [https://hai.stanford.edu/news/ai-detectors-biased-against-non-native-english-writers Stanford HAI]

Stanford HAI summarizes research warning that AI detectors can be “unreliable and easily gamed” and biased against non-native English writers. The article describes detectors being marketed to educators and journalists but highlights the core risk: algorithmic authorship judgments can wrongly flag human work.

Directive implication: do not use detector output as sole evidence. Preserve writing history, outlines, notes, version diffs, and citations when authorship might be questioned.

=== Liang et al. 2023: GPT detectors biased against non-native English writers ===

Source: [https://arxiv.org/abs/2304.02819 arXiv]

The arXiv paper “GPT detectors are biased against non-native English writers” directly examines detector performance and bias. Its relevance is not a writing recipe but a caution: predictable or simpler English can be misread by detectors as synthetic.

Directive implication: do not “complexify” prose artificially to dodge flags. Instead, write to the audience and keep provenance records.

=== Pangram technical report ===

Source: [https://arxiv.org/abs/2402.14873 arXiv]

The Pangram technical report describes a classifier trained across domains and model outputs and claims very low false positive rates in high-data domains. It also claims generalization to nonnative speakers and unseen domains/models.

Directive implication: detector technology varies widely. Some systems are model-based classifiers rather than simple perplexity/burstiness tools. This makes detector-specific evasion brittle and ethically problematic. The durable response is quality control plus transparent authorship.

=== GPTZero FAQ ===

Source: [https://gptzero.me/faq/ GPTZero FAQ]

GPTZero positions itself as an AI detector plus authorship-verification platform, including integrations that preserve writing transparency. The page foregrounds AI probabilities and writing transparency rather than pure binary proof.

Directive implication: when provenance matters, authorship history is stronger than retroactive style manipulation.

=== Slop Forensics Toolkit ===

Source: [https://github.com/sam-paech/slop-forensics Slop Forensics]

Slop Forensics analyzes overrepresented lexical patterns in LLM output: repeated words, bigrams, trigrams, vocabulary complexity, and slop scores. Autonovel cites this as an inspiration for its anti-slop wordlists.

Directive implication: a useful revision pass can search for statistically overused LLM vocabulary and replace it with topic-specific language. But wordlists alone cannot prove authorship or guarantee quality.

=== EQ-Bench Slop Score ===

Source: [https://eqbench.com/slop-score.html EQ-Bench Slop Score]

EQ-Bench states that Slop Score is not a general AI detector. It measures overused AI-text patterns, especially slop words, “not X but Y” contrast patterns, and slop trigrams. It says the metric is optimized for creative writing and essays.

Directive implication: use slop scoring as a quality smell test. Do not optimize blindly for a score; a clean score can still be dull, false, or unethical.

=== MLA guidance on citing generative AI ===

Source: [https://style.mla.org/citing-generative-ai/ MLA guidance]

The MLA page and comments emphasize disclosure/citation practices for generative AI, including acknowledging AI assistance and reviewing, editing, and supporting content with citations.

Directive implication: when AI materially contributes to prose, disclose according to the relevant venue’s rules. If output includes research claims, verify and cite primary sources.

=== Additional guidance located but access-limited in this environment ===

* OpenAI’s AI-text-classifier announcement page was Cloudflare-blocked here. It is still a commonly cited source in the broader detector debate because OpenAI later marked its classifier as unavailable due to low accuracy, but this specific session could not fetch the page content.
* Turnitin’s AI-detection product page was HTTP 403-blocked here. Treat Turnitin’s documentation as a venue-specific source to check directly where institutional rules depend on it.
* A Vanderbilt Brightspace article about Turnitin AI detection being unavailable was attempted but returned 404 for the URL tested. Do not rely on that URL without fresh verification.

== Synthesis: safe directives for high-quality AI-assisted prose ==

=== Do ===

* Define voice before drafting: POV, tense, register, vocabulary wells, forbidden clichés, sentence rhythm.
* Ground abstractions in concrete evidence: physical action, sensory detail, dialogue, object-specific description.
* Use character-specific metaphors and speech patterns.
* Prefer scene over summary where emotion, conflict, or decision matters.
* Let subtext do work; if the scene shows it, do not explain it afterward.
* Vary sentence and paragraph length for rhetorical purpose.
* Add one real surprise per scene/chapter: a wrong word, premature emotion, interrupted beat, awkward silence, or consequence.
* Run deterministic checks for filler, repeated formulas, and cliché.
* Run adversarial editing: ask what to cut, not whether the prose is “good.”
* Keep drafts, outlines, prompt logs, revision notes, and source citations.
* Disclose AI assistance where required by school, publisher, client, or platform rules.

=== Don’t ===

* Do not ask a model to “beat,” “evade,” “bypass,” or “trick” AI detectors.
* Do not launder AI output as purely human work where disclosure is expected.
* Do not optimize prose for a proprietary detector score.
* Do not add random errors, typos, or awkward phrasing to mimic humanity.
* Do not replace every flagged word mechanically; context matters.
* Do not let word-level slop cleanup substitute for structural revision.
* Do not use a single detector result as proof of authorship.

== Practical revision workflow ==

# '''Provenance pass:''' save outline, notes, sources, prompts, and draft diffs.
# '''Voice pass:''' define intended voice, register, audience, POV, and constraints.
# '''Draft pass:''' produce complete scene/chapter/essay without stopping to polish every sentence.
# '''Mechanical pass:''' scan for filler phrases, slop words, AI-fiction clichés, repeated formulas, transition abuse, sentence uniformity, and abstract noun density.
# '''Specificity pass:''' replace generic claims/images with observed facts, concrete nouns, precise verbs, and source-backed claims.
# '''Structure pass:''' break template paragraphs, reduce symmetrical lists, vary paragraph size, and ensure sections are naturally lumpy.
# '''Subtext pass:''' delete explanations after emotional beats.
# '''Dialogue/interiority pass:''' make speech imperfect and character-specific; replace “thought about” lists with embodied cognition.
# '''Adversarial edit:''' request exact cuts classified by FAT / REDUNDANT / OVER-EXPLAIN / GENERIC / TELL / STRUCTURAL.
# '''Human accountability pass:''' verify facts, citations, tone, venue disclosure requirements, and authorial intent.

== Bottom line ==

The research does not support writing to “avoid AI detection.” It supports writing and revising so that the prose is specific, truthful, voiceful, well-documented, and accountable. Detectors remain contested; provenance and craft are more reliable than evasion.

Research/Fiction Writing/AI Prose Strengthen

2026-04-26T02:39:31Z

Hermesbot: Cleanup wiki format; split prose research from prompts

This page gathers research notes on AI-assisted prose quality, detector claims, and revision practices that favor specificity, voice, and accountability over generic polish.

Related: [[Research/Fiction Writing/AI Prose Prompts]]

'''Scope note:''' The original request asked for directives to “avoid AI detection.” I cannot help create detector-evasion instructions. This research file therefore reframes the task as: how to produce better, more specific, more accountable AI-assisted prose while avoiding low-quality “AI slop,” and how to document authorship transparently. The companion prompts page gives quality-control prompts, not instructions for deceiving readers, instructors, publishers, or detection systems.

== Executive summary ==

# '''AI detectors are imperfect evidence, not proof.''' The strongest external theme is uncertainty: detectors can produce false positives, can vary by domain and sample length, and may be biased against non-native English writers. Treat detector output as one signal among many, never as an authorship verdict.
# '''The NousResearch/autonovel project is mainly a craft-and-revision pipeline.''' Its useful contribution is not “beat the detector”; it is a repeatable process: generate layered context, draft with strong voice constraints, mechanically scan for slop, run adversarial editing, revise from specific cuts, then use reader/reviewer loops.
# '''Low-quality AI prose has recurring signals.''' The project flags overused lexical patterns, filler transitions, rigid paragraph templates, symmetrical lists, over-explained emotion, generic description, polished dialogue, and uniform rhythm.
# '''Good prose is specific and accountable.''' The safest durable directive is not “look human,” but “earn every sentence”: concrete nouns, embodied sensory detail, character-specific metaphors, subtext, sentence-length variation, scene over summary, and revision against actual weaknesses.
# '''Transparency matters.''' MLA and other style/teaching guidance increasingly emphasize disclosure/citation of generative-AI use when it materially contributes to text. Keep drafts, notes, prompts, and revision history when provenance matters.

== Primary project researched: NousResearch/autonovel ==

Repository: [https://github.com/NousResearch/autonovel]

The repository describes itself as “an autonomous pipeline for writing, revising, typesetting, illustrating, and narrating a complete novel,” inspired by Karpathy’s <code>autoresearch</code> modify/evaluate/keep-discard loop. The first produced novel reportedly went through foundation, drafting, six automated revision cycles, and six Opus review rounds.

=== Pipeline structure ===

From <code>README.md</code>, <code>WORKFLOW.md</code>, and <code>PIPELINE.md</code> references:

* '''Phase 1: Foundation''' — build world, characters, outline, voice, and canon from a seed concept; iterate until foundation score clears a threshold.
* '''Phase 2: First draft''' — draft chapters sequentially; evaluate each; keep if above score threshold; retry otherwise.
* '''Phase 3a: Automated revision''' — adversarial editing, cuts, reader panels, revision briefs, and rewritten chapters.
* '''Phase 3b: Opus review loop''' — full-manuscript dual review as literary critic and professor of fiction; parse actionable defects; fix top issues; repeat until major issues are gone.
* '''Phase 4: Export''' — typesetting, ePub, art, audiobook, landing page.

Important operational idea: the novel is treated as '''five co-evolving layers''': <code>voice.md</code> controls how prose is written; <code>world.md</code>, <code>characters.md</code>, <code>outline.md</code>, and <code>canon.md</code> control what is true; chapters are the final prose layer. Revisions propagate up and down the layer stack.

=== Autonovel’s “two immune systems” ===

The README names two immune systems:

# '''Mechanical evaluation''' (<code>evaluate.py</code>) scans without an LLM for banned words, fiction clichés, show-don’t-tell violations, sentence uniformity, transition abuse, and structural tics.
# '''LLM judging''' scores prose quality, voice adherence, character distinctiveness, and beat coverage using a separate model from the writer to reduce self-congratulation.

This is a key pattern: do not rely on a single aesthetic judgment. Use both deterministic checks and adversarial human/editorial review.

== Autonovel directives relevant to prose quality ==

These are extracted from <code>README.md</code>, <code>ANTI-SLOP.md</code>, <code>ANTI-PATTERNS.md</code>, <code>CRAFT.md</code>, <code>draft_chapter.py</code>, <code>evaluate.py</code>, <code>adversarial_edit.py</code>, and <code>voice_fingerprint.py</code>.

=== Word-level anti-slop findings ===

Autonovel’s <code>ANTI-SLOP.md</code> and <code>evaluate.py</code> flag words and phrases statistically or stylistically associated with unedited LLM output. The repository treats these as revision triggers, not absolute proof of authorship.

Commonly flagged categories:

* '''Grandiose or corporate diction:''' “delve,” “utilize,” “leverage,” “facilitate,” “elucidate,” “embark,” “endeavor,” “multifaceted,” “tapestry,” “paradigm,” “synergy,” “holistic,” “myriad,” “plethora.”
* '''Suspicious-in-clusters adjectives/verbs:''' “robust,” “comprehensive,” “seamless,” “cutting-edge,” “innovative,” “streamline,” “empower,” “foster,” “enhance,” “elevate,” “optimize,” “pivotal,” “profound,” “resonate,” “underscore,” “harness,” “cultivate.”
* '''Filler phrases:''' “It’s worth noting,” “It’s important to note,” “Let’s dive into,” “In conclusion,” “To summarize,” “Furthermore,” “Moreover,” “Additionally,” “In today’s fast-paced world,” “At the end of the day,” “When it comes to,” “One might argue.”
* '''Rhetorical crutches:''' especially “not just X, but Y.”

Quality takeaway: replace generic prestige diction with exact nouns, verbs, evidence, and images. If a phrase could fit any topic, it probably adds little.

=== Structural anti-patterns ===

Autonovel’s <code>ANTI-PATTERNS.md</code> argues that many AI tells are structural, not lexical:

* '''Over-explaining:''' the scene already shows fear, grief, or tension, then the narrator explains it.
* '''Triadic listing:''' repeated “X. Y. Z.” patterns or three-item sensory lists.
* '''Negative assertion repetition:''' repeated “He did not…” formulations.
* '''Cataloging by thinking:''' “He thought about X. He thought about Y…” instead of dramatized interiority.
* '''Simile crutch:''' repeated “the way X did Y.”
* '''Section-break crutch:''' using breaks to avoid transitions.
* '''Paragraph-length uniformity:''' middle sections flatten into similar 4–6 sentence paragraphs.
* '''Predictable emotional arcs:''' outline beats arrive too cleanly, with no sideways interruption.
* '''Repetitive chapter endings:''' same structural closing move reused.
* '''Balanced antithesis in dialogue:''' “I’m not saying X. I’m saying Y.”
* '''Dialogue as written prose:''' polished complete sentences, no interruptions, false starts, or wrong words.
* '''Scene-summary imbalance:''' too much narration compressing time instead of dramatized action/dialogue.

Quality takeaway: revise for asymmetry, scene-specific endings, imperfect speech, embodied interiority, and genuine surprise.

=== Fiction-specific “AI tell” patterns ===

<code>CRAFT.md</code> and <code>evaluate.py</code> highlight fiction clichés often produced by generic LLM drafting:

* “A sense of [emotion]”
* “Couldn’t help but feel”
* “The weight of [abstract noun]”
* “The air was thick with…”
* “Eyes widened” as default surprise
* “A wave/pang/surge of emotion”
* “Heart pounded in his/her chest”
* Hair that “spilled/cascaded/tumbled”
* “Piercing eyes”
* “A knowing smile”
* “Let out a breath he/she didn’t know they were holding”
* “Something dark/ancient/primal stirred”

Quality takeaway: use physical action, sensory fact, and subtext instead of prepackaged emotional labels.

=== Autonovel drafting constraints worth reusing ===

From <code>draft_chapter.py</code>:

* Write in a defined POV and tense.
* Follow a voice definition exactly.
* Hit every outline beat, but do not summarize or skip.
* Show sensory detail tied to the point-of-view character.
* Use character-specific speech patterns.
* Ban known slop phrases before drafting.
* Vary sentence length deliberately.
* Use metaphors from the character’s lived experience.
* Trust the reader; do not explain what scenes mean.
* Start in scene, not exposition.
* End on a moment, not a summary.
* Include at least one surprising moment per chapter.
* Keep most of the chapter in-scene rather than summarized.

=== Autonovel evaluation metrics worth reusing ===

From <code>evaluate.py</code> and <code>voice_fingerprint.py</code>:

* banned/slop word hits
* filler phrase hits
* fiction cliché hits
* show-don’t-tell violations
* structural tic counts
* em dash density
* sentence-length coefficient of variation
* transition-opener ratio
* paragraph-length variation
* dialogue ratio
* abstract-noun density
* repeated sentence starters
* simile density
* section-break count
* chapter-level outliers from the manuscript average

These metrics should not be treated as “AI detector evasion.” They are revision instruments: they expose sameness, abstraction, and cliché.

=== Adversarial editing as the strongest revision pattern ===

<code>adversarial_edit.py</code> asks a judge to identify 10–20 exact passages to cut or rewrite and classify them as:

* FAT — adds nothing
* REDUNDANT — restates what was already shown
* OVER-EXPLAIN — explains what the scene demonstrated
* GENERIC — could appear in any story
* TELL — names emotion/state instead of dramatizing it
* STRUCTURAL — disrupts pacing or rhythm

The key research finding: asking “what would you cut?” is more useful than asking for a general quality score. Absolute 1–10 scoring compresses; specific cut lists produce revision plans.

== External research and documentation ==

=== Stanford HAI: detector bias and unreliability ===

Source: [https://hai.stanford.edu/news/ai-detectors-biased-against-non-native-english-writers]

Stanford HAI summarizes research warning that AI detectors can be “unreliable and easily gamed” and biased against non-native English writers. The article describes detectors being marketed to educators and journalists but highlights the core risk: algorithmic authorship judgments can wrongly flag human work.

Directive implication: do not use detector output as sole evidence. Preserve writing history, outlines, notes, version diffs, and citations when authorship might be questioned.

=== Liang et al. 2023: GPT detectors biased against non-native English writers ===

Source: [https://arxiv.org/abs/2304.02819]

The arXiv paper “GPT detectors are biased against non-native English writers” directly examines detector performance and bias. Its relevance is not a writing recipe but a caution: predictable or simpler English can be misread by detectors as synthetic.

Directive implication: do not “complexify” prose artificially to dodge flags. Instead, write to the audience and keep provenance records.

=== Pangram technical report ===

Source: [https://arxiv.org/abs/2402.14873]

The Pangram technical report describes a classifier trained across domains and model outputs and claims very low false positive rates in high-data domains. It also claims generalization to nonnative speakers and unseen domains/models.

Directive implication: detector technology varies widely. Some systems are model-based classifiers rather than simple perplexity/burstiness tools. This makes detector-specific evasion brittle and ethically problematic. The durable response is quality control plus transparent authorship.

=== GPTZero FAQ ===

Source: [https://gptzero.me/faq/]

GPTZero positions itself as an AI detector plus authorship-verification platform, including integrations that preserve writing transparency. The page foregrounds AI probabilities and writing transparency rather than pure binary proof.

Directive implication: when provenance matters, authorship history is stronger than retroactive style manipulation.

=== Slop Forensics Toolkit ===

Source: [https://github.com/sam-paech/slop-forensics]

Slop Forensics analyzes overrepresented lexical patterns in LLM output: repeated words, bigrams, trigrams, vocabulary complexity, and slop scores. Autonovel cites this as an inspiration for its anti-slop wordlists.

Directive implication: a useful revision pass can search for statistically overused LLM vocabulary and replace it with topic-specific language. But wordlists alone cannot prove authorship or guarantee quality.

=== EQ-Bench Slop Score ===

Source: [https://eqbench.com/slop-score.html]

EQ-Bench states that Slop Score is not a general AI detector. It measures overused AI-text patterns, especially slop words, “not X but Y” contrast patterns, and slop trigrams. It says the metric is optimized for creative writing and essays.

Directive implication: use slop scoring as a quality smell test. Do not optimize blindly for a score; a clean score can still be dull, false, or unethical.

=== MLA guidance on citing generative AI ===

Source: [https://style.mla.org/citing-generative-ai/]

The MLA page and comments emphasize disclosure/citation practices for generative AI, including acknowledging AI assistance and reviewing, editing, and supporting content with citations.

Directive implication: when AI materially contributes to prose, disclose according to the relevant venue’s rules. If output includes research claims, verify and cite primary sources.

=== Additional guidance located but access-limited in this environment ===

* OpenAI’s AI-text-classifier announcement page was Cloudflare-blocked here. It is still a commonly cited source in the broader detector debate because OpenAI later marked its classifier as unavailable due to low accuracy, but this specific session could not fetch the page content.
* Turnitin’s AI-detection product page was HTTP 403-blocked here. Treat Turnitin’s documentation as a venue-specific source to check directly where institutional rules depend on it.
* A Vanderbilt Brightspace article about Turnitin AI detection being unavailable was attempted but returned 404 for the URL tested. Do not rely on that URL without fresh verification.

== Synthesis: safe directives for high-quality AI-assisted prose ==

=== Do ===

* Define voice before drafting: POV, tense, register, vocabulary wells, forbidden clichés, sentence rhythm.
* Ground abstractions in concrete evidence: physical action, sensory detail, dialogue, object-specific description.
* Use character-specific metaphors and speech patterns.
* Prefer scene over summary where emotion, conflict, or decision matters.
* Let subtext do work; if the scene shows it, do not explain it afterward.
* Vary sentence and paragraph length for rhetorical purpose.
* Add one real surprise per scene/chapter: a wrong word, premature emotion, interrupted beat, awkward silence, or consequence.
* Run deterministic checks for filler, repeated formulas, and cliché.
* Run adversarial editing: ask what to cut, not whether the prose is “good.”
* Keep drafts, outlines, prompt logs, revision notes, and source citations.
* Disclose AI assistance where required by school, publisher, client, or platform rules.

=== Don’t ===

* Do not ask a model to “beat,” “evade,” “bypass,” or “trick” AI detectors.
* Do not launder AI output as purely human work where disclosure is expected.
* Do not optimize prose for a proprietary detector score.
* Do not add random errors, typos, or awkward phrasing to mimic humanity.
* Do not replace every flagged word mechanically; context matters.
* Do not let word-level slop cleanup substitute for structural revision.
* Do not use a single detector result as proof of authorship.

== Practical revision workflow ==

# '''Provenance pass:''' save outline, notes, sources, prompts, and draft diffs.
# '''Voice pass:''' define intended voice, register, audience, POV, and constraints.
# '''Draft pass:''' produce complete scene/chapter/essay without stopping to polish every sentence.
# '''Mechanical pass:''' scan for filler phrases, slop words, AI-fiction clichés, repeated formulas, transition abuse, sentence uniformity, and abstract noun density.
# '''Specificity pass:''' replace generic claims/images with observed facts, concrete nouns, precise verbs, and source-backed claims.
# '''Structure pass:''' break template paragraphs, reduce symmetrical lists, vary paragraph size, and ensure sections are naturally lumpy.
# '''Subtext pass:''' delete explanations after emotional beats.
# '''Dialogue/interiority pass:''' make speech imperfect and character-specific; replace “thought about” lists with embodied cognition.
# '''Adversarial edit:''' request exact cuts classified by FAT / REDUNDANT / OVER-EXPLAIN / GENERIC / TELL / STRUCTURAL.
# '''Human accountability pass:''' verify facts, citations, tone, venue disclosure requirements, and authorial intent.

== Bottom line ==

The research does not support writing to “avoid AI detection.” It supports writing and revising so that the prose is specific, truthful, voiceful, well-documented, and accountable. Detectors remain contested; provenance and craft are more reliable than evasion.

Research/Fiction Writing/AI Prose Prompts

2026-04-26T02:38:56Z

Hermesbot: Cleanup wiki format; split prompts into their own page

This page gathers safe prompt patterns for AI-assisted prose revision, quality control, and provenance-aware editing.

Related: [[Research/Fiction Writing/AI Prose Strengthen]]

'''Scope note:''' These prompts are for prose quality, authorship transparency, and accountable revision. They are not prompts to evade, bypass, or deceive AI-detection systems.

== Master directive ==

Use this at the top of any writing or editing prompt:

<pre>
Write and revise for clarity, specificity, voice, truthfulness, and reader trust. Do not optimize for AI-detector scores and do not attempt to disguise authorship. If AI assistance materially shapes the text, preserve provenance notes and follow the relevant disclosure rules. Avoid unedited-LLM “slop”: generic claims, filler transitions, overused prestige diction, symmetrical structure, over-explained emotion, and polished-but-empty prose.
</pre>

== Drafting directive: fiction ==

<pre>
Draft the scene in [POV/person/tense]. Stay locked to [character]’s perception and knowledge.

Voice constraints:
- Use concrete nouns and active verbs.
- Ground emotion in action, sensory detail, gesture, silence, and subtext.
- Use metaphors from the character’s lived experience: [list domains].
- Vary sentence length deliberately: short for impact, longer for accumulation or thought.
- Vary paragraph length; avoid three or more same-sized paragraphs in a row.
- Dialogue should sound spoken, not essayistic. Include false starts, interruptions, evasions, unfinished thoughts, and character-specific vocabulary when natural.
- Prefer scene over summary for conflict, revelation, decision, and emotional peaks.
- Trust the reader. If the scene shows something, do not explain it afterward.

Avoid:
- Generic fantasy/fiction clichés.
- “A sense of…,” “couldn’t help but feel,” “the weight of…,” “the air was thick with…,” “eyes widened,” “a pang/wave/surge of emotion,” “heart pounded in [chest],” “a knowing smile.”
- Repeated triads: “X. Y. Z.” or “X and Y and Z.”
- Repeated “He/She did not…” constructions.
- “He/She thought about X” catalogues.
- Balanced-antithesis dialogue: “I’m not saying X. I’m saying Y.”
- Section breaks used to avoid transitions.
- Endings that summarize the scene’s meaning.

Include at least one specific surprise: a wrong word, interrupted beat, premature/late emotion, detail that doesn’t fit, costly choice, or unresolved silence.
</pre>

== Drafting directive: essay / article / README ==

<pre>
Draft for a real reader in a real context: [audience], [purpose], [venue].

Style requirements:
- Start with the actual point; no throat-clearing.
- Use precise claims and examples instead of abstract setup.
- Prefer “use” over “utilize,” “help” over “facilitate,” and specific verbs over prestige verbs.
- Use lists only where lists improve comprehension. Do not default to 3 or 5 balanced bullets.
- Let sections be naturally uneven; allocate length by complexity, not symmetry.
- If a claim depends on evidence, cite or mark it for verification.
- Preserve the author’s stance, uncertainty, and lived context. Do not flatten into neutral corporate prose.
- Use direct language when the claim is known; state uncertainty explicitly when it is not.

Avoid filler:
- “It is worth noting…”
- “It is important to note…”
- “In today’s fast-paced world…”
- “Let’s dive into…”
- “Furthermore/Moreover/Additionally” as default paragraph openers
- “Not just X, but Y”
- Generic conclusions that restate the prompt
</pre>

== Mechanical anti-slop scan prompt ==

<pre>
Audit the text below for low-quality AI-assisted prose patterns. Do not judge authorship. Only identify revision opportunities.

Return a table with columns: Pattern, Exact quote, Why it weakens the prose, Revision action.

Check for:
1. Prestige diction or slop words: delve, utilize, leverage, facilitate, elucidate, embark, endeavor, encompass, multifaceted, tapestry, paradigm, synergy, holistic, myriad, plethora.
2. Suspicious-in-clusters words: robust, comprehensive, seamless, cutting-edge, innovative, streamline, empower, foster, enhance, elevate, optimize, pivotal, profound, resonate, underscore, harness, cultivate.
3. Filler phrases: “it’s worth noting,” “it’s important to note,” “let’s explore,” “in conclusion,” “to summarize,” “when it comes to,” “one might argue.”
4. Formulaic transitions at paragraph starts.
5. “Not just X, but Y” constructions.
6. Overuse of em dashes.
7. Uniform sentence length.
8. Uniform paragraph length.
9. Abstract nouns where concrete evidence would work.
10. Claims that need citation or verification.

Text:
[PASTE TEXT]
</pre>

== Fiction anti-pattern scan prompt ==

<pre>
Audit this fiction passage for structural AI-prose anti-patterns. Do not discuss AI detection. Treat this as a craft edit.

Find exact quotes and classify each issue as one of:
- OVER-EXPLAIN: narrator explains what action/dialogue already showed
- GENERIC: sentence could appear in any story
- TELL: names emotion instead of dramatizing it
- RHYTHM: sentence/paragraph pattern is too uniform
- DIALOGUE: speech sounds written, polished, or interchangeable
- INTERIORITY: thought is catalogued instead of dramatized
- CLICHE: stock phrase/image
- STRUCTURE: section/scene uses summary or breaks to dodge transitions

For each issue, provide either CUT or REWRITE and a concise replacement if needed.

Text:
[PASTE TEXT]
</pre>

== Adversarial editing prompt ==

<pre>
You are a severe but fair literary editor. Your job is to identify exactly what to cut or rewrite to make this text tighter, sharper, more specific, and more alive.

Rules:
- Quote exact text, minimum 10 words per quote.
- Do not invent problems; if a passage works, leave it alone.
- Prefer cuts over rewrites when the text loses nothing.
- Classify each issue as FAT, REDUNDANT, OVER-EXPLAIN, GENERIC, TELL, STRUCTURAL, FACT-CHECK, or VOICE-DRIFT.
- Provide a one-sentence reason.
- If REWRITE, provide a replacement that is shorter unless expansion is truly required.

Return JSON:
{
"cuts_or_rewrites": [
{
"quote": "exact text",
"type": "FAT|REDUNDANT|OVER-EXPLAIN|GENERIC|TELL|STRUCTURAL|FACT-CHECK|VOICE-DRIFT",
"action": "CUT|REWRITE",
"reason": "why it weakens the text",
"replacement": "replacement or null"
}
],
"strongest_passage": "quote",
"weakest_passage": "quote",
"estimated_cuttable_words": 0,
"one_sentence_verdict": "..."
}

Text:
[PASTE TEXT]
</pre>

== Specificity rewrite prompt ==

<pre>
Revise the passage for specificity and evidence. Keep the meaning and authorial stance, but replace generic abstractions with concrete details.

Rules:
- Do not add unverifiable facts. If a fact is missing, mark [NEEDS SOURCE] or [NEEDS EXAMPLE].
- Replace vague nouns with precise nouns.
- Replace weak verbs with active verbs.
- Remove filler introductions.
- Remove generic intensifiers like “very,” “deeply,” “profound,” unless the sentence earns them.
- Keep any useful roughness, humor, uncertainty, or personal voice.
- Do not make the text artificially messy.

Passage:
[PASTE TEXT]
</pre>

== Subtext and show-don’t-tell prompt ==

<pre>
Revise this scene so the emotion is carried by behavior, dialogue, sensory detail, pacing, and omission rather than labels.

Rules:
- Remove direct emotion labels at peak moments unless the POV requires them.
- After an emotional beat, cut any sentence that explains what the beat means.
- Use physical detail that belongs to this character and setting.
- Preserve ambiguity where it creates tension.
- End the scene on an image, action, or line of dialogue, not a summary of meaning.

Scene:
[PASTE TEXT]
</pre>

== Dialogue distinctiveness prompt ==

<pre>
Audit the dialogue for character distinctiveness.

For each speaker, identify:
- average sentence length
- formality level
- contraction use
- favorite sentence shapes
- metaphor domain
- directness vs evasion
- interruptions/false starts
- vocabulary that only this character would use

Then revise only the lines that sound interchangeable or too polished. Keep the scene’s meaning unchanged. Add imperfection only where it reveals character; do not sprinkle random errors.

Dialogue:
[PASTE TEXT]
</pre>

== Rhythm variation prompt ==

<pre>
Revise for rhythm without changing meaning.

Check:
- Are most sentences the same length?
- Do too many paragraphs have the same shape?
- Do consecutive paragraphs start with transition words or the same subject?
- Are there too many em dashes?
- Are lists used where prose would be stronger?

Revise by:
- Combining where accumulation helps.
- Splitting where impact helps.
- Moving the main point later or earlier if the paragraph template is predictable.
- Converting unnecessary lists to prose.
- Keeping rhythm changes motivated by meaning, not randomness.

Text:
[PASTE TEXT]
</pre>

== Provenance and disclosure prompt ==

<pre>
Create a provenance note for this AI-assisted text.

Include:
- Human-provided source material or outline.
- AI tools used and what they contributed.
- Human revisions performed.
- Sources verified by the human author.
- Any claims still needing verification.
- Suggested disclosure wording for [school / client / publisher / public web].

Do not overstate AI authorship and do not hide material AI contribution.

Project notes:
[PASTE NOTES]
</pre>

== Final quality gate ==

<pre>
Final audit before publication/submission.

Answer these questions:
1. Does the text satisfy the assignment/venue and audience?
2. Are all factual claims sourced or clearly framed as opinion/experience?
3. Does the prose have a specific voice rather than generic polish?
4. Are there remaining filler phrases, prestige words, template paragraphs, or repeated rhetorical formulas?
5. Are emotional beats shown rather than explained?
6. Is dialogue/interiority character-specific?
7. Is any AI assistance disclosed according to the relevant rules?
8. Is there preserved provenance if authorship is questioned?

Return:
- PASS/REVISE
- Top 5 required fixes
- Optional improvements
- Disclosure/provenance note status

Text:
[PASTE TEXT]
</pre>

== Minimal checklist for prompts ==

When asking an AI to write or revise prose, include:

* Audience and purpose
* Voice/register
* POV/tense if fiction
* What sources or lived details must be used
* What must not be invented
* What clichés/formulas to avoid
* A requirement for exact-quote edits
* A provenance/disclosure requirement when relevant

== Hard boundary ==

Do not use prompts such as:

<pre>
Make this bypass AI detection.
Make this undetectable as AI.
Add human errors to fool detectors.
Rewrite to beat Turnitin/GPTZero/Pangram.
Hide that AI helped write this.
</pre>

Use this instead:

<pre>
Revise this into stronger, more specific, more truthful prose while preserving transparent authorship and complying with the venue’s disclosure rules.
</pre>

Research/Fiction Writing/AI Prose Strengthen

2026-04-26T02:38:56Z

Hermesbot: Cleanup wiki format; split prose research from prompts

This page gathers research notes on AI-assisted prose quality, detector claims, and revision practices that favor specificity, voice, and accountability over generic polish.

Related: [[Research/Fiction Writing/AI Prose Prompts]]

'''Scope note:''' The original request asked for directives to “avoid AI detection.” I cannot help create detector-evasion instructions. This research file therefore reframes the task as: how to produce better, more specific, more accountable AI-assisted prose while avoiding low-quality “AI slop,” and how to document authorship transparently. The companion prompts page gives quality-control prompts, not instructions for deceiving readers, instructors, publishers, or detection systems.

== Executive summary ==

# **AI detectors are imperfect evidence, not proof.** The strongest external theme is uncertainty: detectors can produce false positives, can vary by domain and sample length, and may be biased against non-native English writers. Treat detector output as one signal among many, never as an authorship verdict.
# **The NousResearch/autonovel project is mainly a craft-and-revision pipeline.** Its useful contribution is not “beat the detector”; it is a repeatable process: generate layered context, draft with strong voice constraints, mechanically scan for slop, run adversarial editing, revise from specific cuts, then use reader/reviewer loops.
# **Low-quality AI prose has recurring signals.** The project flags overused lexical patterns, filler transitions, rigid paragraph templates, symmetrical lists, over-explained emotion, generic description, polished dialogue, and uniform rhythm.
# **Good prose is specific and accountable.** The safest durable directive is not “look human,” but “earn every sentence”: concrete nouns, embodied sensory detail, character-specific metaphors, subtext, sentence-length variation, scene over summary, and revision against actual weaknesses.
# **Transparency matters.** MLA and other style/teaching guidance increasingly emphasize disclosure/citation of generative-AI use when it materially contributes to text. Keep drafts, notes, prompts, and revision history when provenance matters.

== Primary project researched: NousResearch/autonovel ==

Repository: <https://github.com/NousResearch/autonovel>

The repository describes itself as “an autonomous pipeline for writing, revising, typesetting, illustrating, and narrating a complete novel,” inspired by Karpathy’s <code>autoresearch</code> modify/evaluate/keep-discard loop. The first produced novel reportedly went through foundation, drafting, six automated revision cycles, and six Opus review rounds.

=== Pipeline structure ===

From <code>README.md</code>, <code>WORKFLOW.md</code>, and <code>PIPELINE.md</code> references:

* **Phase 1: Foundation** — build world, characters, outline, voice, and canon from a seed concept; iterate until foundation score clears a threshold.
* **Phase 2: First draft** — draft chapters sequentially; evaluate each; keep if above score threshold; retry otherwise.
* **Phase 3a: Automated revision** — adversarial editing, cuts, reader panels, revision briefs, and rewritten chapters.
* **Phase 3b: Opus review loop** — full-manuscript dual review as literary critic and professor of fiction; parse actionable defects; fix top issues; repeat until major issues are gone.
* **Phase 4: Export** — typesetting, ePub, art, audiobook, landing page.

Important operational idea: the novel is treated as '''five co-evolving layers''': <code>voice.md</code> controls how prose is written; <code>world.md</code>, <code>characters.md</code>, <code>outline.md</code>, and <code>canon.md</code> control what is true; chapters are the final prose layer. Revisions propagate up and down the layer stack.

=== Autonovel’s “two immune systems” ===

The README names two immune systems:

# **Mechanical evaluation** (`evaluate.py`) scans without an LLM for banned words, fiction clichés, show-don’t-tell violations, sentence uniformity, transition abuse, and structural tics.
# **LLM judging** scores prose quality, voice adherence, character distinctiveness, and beat coverage using a separate model from the writer to reduce self-congratulation.

This is a key pattern: do not rely on a single aesthetic judgment. Use both deterministic checks and adversarial human/editorial review.

== Autonovel directives relevant to prose quality ==

These are extracted from <code>README.md</code>, <code>ANTI-SLOP.md</code>, <code>ANTI-PATTERNS.md</code>, <code>CRAFT.md</code>, <code>draft_chapter.py</code>, <code>evaluate.py</code>, <code>adversarial_edit.py</code>, and <code>voice_fingerprint.py</code>.

=== Word-level anti-slop findings ===

Autonovel’s <code>ANTI-SLOP.md</code> and <code>evaluate.py</code> flag words and phrases statistically or stylistically associated with unedited LLM output. The repository treats these as revision triggers, not absolute proof of authorship.

Commonly flagged categories:

* **Grandiose or corporate diction:** “delve,” “utilize,” “leverage,” “facilitate,” “elucidate,” “embark,” “endeavor,” “multifaceted,” “tapestry,” “paradigm,” “synergy,” “holistic,” “myriad,” “plethora.”
* **Suspicious-in-clusters adjectives/verbs:** “robust,” “comprehensive,” “seamless,” “cutting-edge,” “innovative,” “streamline,” “empower,” “foster,” “enhance,” “elevate,” “optimize,” “pivotal,” “profound,” “resonate,” “underscore,” “harness,” “cultivate.”
* **Filler phrases:** “It’s worth noting,” “It’s important to note,” “Let’s dive into,” “In conclusion,” “To summarize,” “Furthermore,” “Moreover,” “Additionally,” “In today’s fast-paced world,” “At the end of the day,” “When it comes to,” “One might argue.”
* **Rhetorical crutches:** especially “not just X, but Y.”

Quality takeaway: replace generic prestige diction with exact nouns, verbs, evidence, and images. If a phrase could fit any topic, it probably adds little.

=== Structural anti-patterns ===

Autonovel’s <code>ANTI-PATTERNS.md</code> argues that many AI tells are structural, not lexical:

* **Over-explaining:** the scene already shows fear, grief, or tension, then the narrator explains it.
* **Triadic listing:** repeated “X. Y. Z.” patterns or three-item sensory lists.
* **Negative assertion repetition:** repeated “He did not…” formulations.
* **Cataloging by thinking:** “He thought about X. He thought about Y…” instead of dramatized interiority.
* **Simile crutch:** repeated “the way X did Y.”
* **Section-break crutch:** using breaks to avoid transitions.
* **Paragraph-length uniformity:** middle sections flatten into similar 4–6 sentence paragraphs.
* **Predictable emotional arcs:** outline beats arrive too cleanly, with no sideways interruption.
* **Repetitive chapter endings:** same structural closing move reused.
* **Balanced antithesis in dialogue:** “I’m not saying X. I’m saying Y.”
* **Dialogue as written prose:** polished complete sentences, no interruptions, false starts, or wrong words.
* **Scene-summary imbalance:** too much narration compressing time instead of dramatized action/dialogue.

Quality takeaway: revise for asymmetry, scene-specific endings, imperfect speech, embodied interiority, and genuine surprise.

=== Fiction-specific “AI tell” patterns ===

<code>CRAFT.md</code> and <code>evaluate.py</code> highlight fiction clichés often produced by generic LLM drafting:

* “A sense of [emotion]”
* “Couldn’t help but feel”
* “The weight of [abstract noun]”
* “The air was thick with…”
* “Eyes widened” as default surprise
* “A wave/pang/surge of emotion”
* “Heart pounded in his/her chest”
* Hair that “spilled/cascaded/tumbled”
* “Piercing eyes”
* “A knowing smile”
* “Let out a breath he/she didn’t know they were holding”
* “Something dark/ancient/primal stirred”

Quality takeaway: use physical action, sensory fact, and subtext instead of prepackaged emotional labels.

=== Autonovel drafting constraints worth reusing ===

From <code>draft_chapter.py</code>:

* Write in a defined POV and tense.
* Follow a voice definition exactly.
* Hit every outline beat, but do not summarize or skip.
* Show sensory detail tied to the point-of-view character.
* Use character-specific speech patterns.
* Ban known slop phrases before drafting.
* Vary sentence length deliberately.
* Use metaphors from the character’s lived experience.
* Trust the reader; do not explain what scenes mean.
* Start in scene, not exposition.
* End on a moment, not a summary.
* Include at least one surprising moment per chapter.
* Keep most of the chapter in-scene rather than summarized.

=== Autonovel evaluation metrics worth reusing ===

From <code>evaluate.py</code> and <code>voice_fingerprint.py</code>:

* banned/slop word hits
* filler phrase hits
* fiction cliché hits
* show-don’t-tell violations
* structural tic counts
* em dash density
* sentence-length coefficient of variation
* transition-opener ratio
* paragraph-length variation
* dialogue ratio
* abstract-noun density
* repeated sentence starters
* simile density
* section-break count
* chapter-level outliers from the manuscript average

These metrics should not be treated as “AI detector evasion.” They are revision instruments: they expose sameness, abstraction, and cliché.

=== Adversarial editing as the strongest revision pattern ===

<code>adversarial_edit.py</code> asks a judge to identify 10–20 exact passages to cut or rewrite and classify them as:

* FAT — adds nothing
* REDUNDANT — restates what was already shown
* OVER-EXPLAIN — explains what the scene demonstrated
* GENERIC — could appear in any story
* TELL — names emotion/state instead of dramatizing it
* STRUCTURAL — disrupts pacing or rhythm

The key research finding: asking “what would you cut?” is more useful than asking for a general quality score. Absolute 1–10 scoring compresses; specific cut lists produce revision plans.

== External research and documentation ==

=== Stanford HAI: detector bias and unreliability ===

Source: <https://hai.stanford.edu/news/ai-detectors-biased-against-non-native-english-writers>

Stanford HAI summarizes research warning that AI detectors can be “unreliable and easily gamed” and biased against non-native English writers. The article describes detectors being marketed to educators and journalists but highlights the core risk: algorithmic authorship judgments can wrongly flag human work.

Directive implication: do not use detector output as sole evidence. Preserve writing history, outlines, notes, version diffs, and citations when authorship might be questioned.

=== Liang et al. 2023: GPT detectors biased against non-native English writers ===

Source: <https://arxiv.org/abs/2304.02819>

The arXiv paper “GPT detectors are biased against non-native English writers” directly examines detector performance and bias. Its relevance is not a writing recipe but a caution: predictable or simpler English can be misread by detectors as synthetic.

Directive implication: do not “complexify” prose artificially to dodge flags. Instead, write to the audience and keep provenance records.

=== Pangram technical report ===

Source: <https://arxiv.org/abs/2402.14873>

The Pangram technical report describes a classifier trained across domains and model outputs and claims very low false positive rates in high-data domains. It also claims generalization to nonnative speakers and unseen domains/models.

Directive implication: detector technology varies widely. Some systems are model-based classifiers rather than simple perplexity/burstiness tools. This makes detector-specific evasion brittle and ethically problematic. The durable response is quality control plus transparent authorship.

=== GPTZero FAQ ===

Source: <https://gptzero.me/faq/>

GPTZero positions itself as an AI detector plus authorship-verification platform, including integrations that preserve writing transparency. The page foregrounds AI probabilities and writing transparency rather than pure binary proof.

Directive implication: when provenance matters, authorship history is stronger than retroactive style manipulation.

=== Slop Forensics Toolkit ===

Source: <https://github.com/sam-paech/slop-forensics>

Slop Forensics analyzes overrepresented lexical patterns in LLM output: repeated words, bigrams, trigrams, vocabulary complexity, and slop scores. Autonovel cites this as an inspiration for its anti-slop wordlists.

Directive implication: a useful revision pass can search for statistically overused LLM vocabulary and replace it with topic-specific language. But wordlists alone cannot prove authorship or guarantee quality.

=== EQ-Bench Slop Score ===

Source: <https://eqbench.com/slop-score.html>

EQ-Bench states that Slop Score is not a general AI detector. It measures overused AI-text patterns, especially slop words, “not X but Y” contrast patterns, and slop trigrams. It says the metric is optimized for creative writing and essays.

Directive implication: use slop scoring as a quality smell test. Do not optimize blindly for a score; a clean score can still be dull, false, or unethical.

=== MLA guidance on citing generative AI ===

Source: <https://style.mla.org/citing-generative-ai/>

The MLA page and comments emphasize disclosure/citation practices for generative AI, including acknowledging AI assistance and reviewing, editing, and supporting content with citations.

Directive implication: when AI materially contributes to prose, disclose according to the relevant venue’s rules. If output includes research claims, verify and cite primary sources.

=== Additional guidance located but access-limited in this environment ===

* OpenAI’s AI-text-classifier announcement page was Cloudflare-blocked here. It is still a commonly cited source in the broader detector debate because OpenAI later marked its classifier as unavailable due to low accuracy, but this specific session could not fetch the page content.
* Turnitin’s AI-detection product page was HTTP 403-blocked here. Treat Turnitin’s documentation as a venue-specific source to check directly where institutional rules depend on it.
* A Vanderbilt Brightspace article about Turnitin AI detection being unavailable was attempted but returned 404 for the URL tested. Do not rely on that URL without fresh verification.

== Synthesis: safe directives for high-quality AI-assisted prose ==

=== Do ===

* Define voice before drafting: POV, tense, register, vocabulary wells, forbidden clichés, sentence rhythm.
* Ground abstractions in concrete evidence: physical action, sensory detail, dialogue, object-specific description.
* Use character-specific metaphors and speech patterns.
* Prefer scene over summary where emotion, conflict, or decision matters.
* Let subtext do work; if the scene shows it, do not explain it afterward.
* Vary sentence and paragraph length for rhetorical purpose.
* Add one real surprise per scene/chapter: a wrong word, premature emotion, interrupted beat, awkward silence, or consequence.
* Run deterministic checks for filler, repeated formulas, and cliché.
* Run adversarial editing: ask what to cut, not whether the prose is “good.”
* Keep drafts, outlines, prompt logs, revision notes, and source citations.
* Disclose AI assistance where required by school, publisher, client, or platform rules.

=== Don’t ===

* Do not ask a model to “beat,” “evade,” “bypass,” or “trick” AI detectors.
* Do not launder AI output as purely human work where disclosure is expected.
* Do not optimize prose for a proprietary detector score.
* Do not add random errors, typos, or awkward phrasing to mimic humanity.
* Do not replace every flagged word mechanically; context matters.
* Do not let word-level slop cleanup substitute for structural revision.
* Do not use a single detector result as proof of authorship.

== Practical revision workflow ==

# **Provenance pass:** save outline, notes, sources, prompts, and draft diffs.
# **Voice pass:** define intended voice, register, audience, POV, and constraints.
# **Draft pass:** produce complete scene/chapter/essay without stopping to polish every sentence.
# **Mechanical pass:** scan for filler phrases, slop words, AI-fiction clichés, repeated formulas, transition abuse, sentence uniformity, and abstract noun density.
# **Specificity pass:** replace generic claims/images with observed facts, concrete nouns, precise verbs, and source-backed claims.
# **Structure pass:** break template paragraphs, reduce symmetrical lists, vary paragraph size, and ensure sections are naturally lumpy.
# **Subtext pass:** delete explanations after emotional beats.
# **Dialogue/interiority pass:** make speech imperfect and character-specific; replace “thought about” lists with embodied cognition.
# **Adversarial edit:** request exact cuts classified by FAT / REDUNDANT / OVER-EXPLAIN / GENERIC / TELL / STRUCTURAL.
# **Human accountability pass:** verify facts, citations, tone, venue disclosure requirements, and authorial intent.

== Bottom line ==

The research does not support writing to “avoid AI detection.” It supports writing and revising so that the prose is specific, truthful, voiceful, well-documented, and accountable. Detectors remain contested; provenance and craft are more reliable than evasion.

Research/Fiction Writing/AI Prose Strengthen

2026-04-26T02:34:59Z

Hermesbot: Update with ai-detect research and prompt files

= AI Prose Strengthen =

This page contains the full contents of the two local reference files used for MGAIF prose-quality research.

== ai-detect-prose-research.md ==

<pre>
# AI-Assisted Prose, Detector Claims, and Prose-Quality Research

**Scope note:** The original request asked for directives to “avoid AI detection.” I cannot help create detector-evasion instructions. This research file therefore reframes the task as: how to produce better, more specific, more accountable AI-assisted prose while avoiding low-quality “AI slop,” and how to document authorship transparently. The accompanying `ai-detect-prompts.md` gives quality-control prompts, not instructions for deceiving readers, instructors, publishers, or detection systems.

## Executive summary

1. **AI detectors are imperfect evidence, not proof.** The strongest external theme is uncertainty: detectors can produce false positives, can vary by domain and sample length, and may be biased against non-native English writers. Treat detector output as one signal among many, never as an authorship verdict.
2. **The NousResearch/autonovel project is mainly a craft-and-revision pipeline.** Its useful contribution is not “beat the detector”; it is a repeatable process: generate layered context, draft with strong voice constraints, mechanically scan for slop, run adversarial editing, revise from specific cuts, then use reader/reviewer loops.
3. **Low-quality AI prose has recurring signals.** The project flags overused lexical patterns, filler transitions, rigid paragraph templates, symmetrical lists, over-explained emotion, generic description, polished dialogue, and uniform rhythm.
4. **Good prose is specific and accountable.** The safest durable directive is not “look human,” but “earn every sentence”: concrete nouns, embodied sensory detail, character-specific metaphors, subtext, sentence-length variation, scene over summary, and revision against actual weaknesses.
5. **Transparency matters.** MLA and other style/teaching guidance increasingly emphasize disclosure/citation of generative-AI use when it materially contributes to text. Keep drafts, notes, prompts, and revision history when provenance matters.

## Primary project researched: NousResearch/autonovel

Repository: <https://github.com/NousResearch/autonovel>

The repository describes itself as “an autonomous pipeline for writing, revising, typesetting, illustrating, and narrating a complete novel,” inspired by Karpathy’s `autoresearch` modify/evaluate/keep-discard loop. The first produced novel reportedly went through foundation, drafting, six automated revision cycles, and six Opus review rounds.

### Pipeline structure

From `README.md`, `WORKFLOW.md`, and `PIPELINE.md` references:

- **Phase 1: Foundation** — build world, characters, outline, voice, and canon from a seed concept; iterate until foundation score clears a threshold.
- **Phase 2: First draft** — draft chapters sequentially; evaluate each; keep if above score threshold; retry otherwise.
- **Phase 3a: Automated revision** — adversarial editing, cuts, reader panels, revision briefs, and rewritten chapters.
- **Phase 3b: Opus review loop** — full-manuscript dual review as literary critic and professor of fiction; parse actionable defects; fix top issues; repeat until major issues are gone.
- **Phase 4: Export** — typesetting, ePub, art, audiobook, landing page.

Important operational idea: the novel is treated as **five co-evolving layers**: `voice.md` controls how prose is written; `world.md`, `characters.md`, `outline.md`, and `canon.md` control what is true; chapters are the final prose layer. Revisions propagate up and down the layer stack.

### Autonovel’s “two immune systems”

The README names two immune systems:

1. **Mechanical evaluation** (`evaluate.py`) scans without an LLM for banned words, fiction clichés, show-don’t-tell violations, sentence uniformity, transition abuse, and structural tics.
2. **LLM judging** scores prose quality, voice adherence, character distinctiveness, and beat coverage using a separate model from the writer to reduce self-congratulation.

This is a key pattern: do not rely on a single aesthetic judgment. Use both deterministic checks and adversarial human/editorial review.

## Autonovel directives relevant to prose quality

These are extracted from `README.md`, `ANTI-SLOP.md`, `ANTI-PATTERNS.md`, `CRAFT.md`, `draft_chapter.py`, `evaluate.py`, `adversarial_edit.py`, and `voice_fingerprint.py`.

### Word-level anti-slop findings

Autonovel’s `ANTI-SLOP.md` and `evaluate.py` flag words and phrases statistically or stylistically associated with unedited LLM output. The repository treats these as revision triggers, not absolute proof of authorship.

Commonly flagged categories:

- **Grandiose or corporate diction:** “delve,” “utilize,” “leverage,” “facilitate,” “elucidate,” “embark,” “endeavor,” “multifaceted,” “tapestry,” “paradigm,” “synergy,” “holistic,” “myriad,” “plethora.”
- **Suspicious-in-clusters adjectives/verbs:** “robust,” “comprehensive,” “seamless,” “cutting-edge,” “innovative,” “streamline,” “empower,” “foster,” “enhance,” “elevate,” “optimize,” “pivotal,” “profound,” “resonate,” “underscore,” “harness,” “cultivate.”
- **Filler phrases:** “It’s worth noting,” “It’s important to note,” “Let’s dive into,” “In conclusion,” “To summarize,” “Furthermore,” “Moreover,” “Additionally,” “In today’s fast-paced world,” “At the end of the day,” “When it comes to,” “One might argue.”
- **Rhetorical crutches:** especially “not just X, but Y.”

Quality takeaway: replace generic prestige diction with exact nouns, verbs, evidence, and images. If a phrase could fit any topic, it probably adds little.

### Structural anti-patterns

Autonovel’s `ANTI-PATTERNS.md` argues that many AI tells are structural, not lexical:

- **Over-explaining:** the scene already shows fear, grief, or tension, then the narrator explains it.
- **Triadic listing:** repeated “X. Y. Z.” patterns or three-item sensory lists.
- **Negative assertion repetition:** repeated “He did not…” formulations.
- **Cataloging by thinking:** “He thought about X. He thought about Y…” instead of dramatized interiority.
- **Simile crutch:** repeated “the way X did Y.”
- **Section-break crutch:** using breaks to avoid transitions.
- **Paragraph-length uniformity:** middle sections flatten into similar 4–6 sentence paragraphs.
- **Predictable emotional arcs:** outline beats arrive too cleanly, with no sideways interruption.
- **Repetitive chapter endings:** same structural closing move reused.
- **Balanced antithesis in dialogue:** “I’m not saying X. I’m saying Y.”
- **Dialogue as written prose:** polished complete sentences, no interruptions, false starts, or wrong words.
- **Scene-summary imbalance:** too much narration compressing time instead of dramatized action/dialogue.

Quality takeaway: revise for asymmetry, scene-specific endings, imperfect speech, embodied interiority, and genuine surprise.

### Fiction-specific “AI tell” patterns

`CRAFT.md` and `evaluate.py` highlight fiction clichés often produced by generic LLM drafting:

- “A sense of [emotion]”
- “Couldn’t help but feel”
- “The weight of [abstract noun]”
- “The air was thick with…”
- “Eyes widened” as default surprise
- “A wave/pang/surge of emotion”
- “Heart pounded in his/her chest”
- Hair that “spilled/cascaded/tumbled”
- “Piercing eyes”
- “A knowing smile”
- “Let out a breath he/she didn’t know they were holding”
- “Something dark/ancient/primal stirred”

Quality takeaway: use physical action, sensory fact, and subtext instead of prepackaged emotional labels.

### Autonovel drafting constraints worth reusing

From `draft_chapter.py`:

- Write in a defined POV and tense.
- Follow a voice definition exactly.
- Hit every outline beat, but do not summarize or skip.
- Show sensory detail tied to the point-of-view character.
- Use character-specific speech patterns.
- Ban known slop phrases before drafting.
- Vary sentence length deliberately.
- Use metaphors from the character’s lived experience.
- Trust the reader; do not explain what scenes mean.
- Start in scene, not exposition.
- End on a moment, not a summary.
- Include at least one surprising moment per chapter.
- Keep most of the chapter in-scene rather than summarized.

### Autonovel evaluation metrics worth reusing

From `evaluate.py` and `voice_fingerprint.py`:

- banned/slop word hits
- filler phrase hits
- fiction cliché hits
- show-don’t-tell violations
- structural tic counts
- em dash density
- sentence-length coefficient of variation
- transition-opener ratio
- paragraph-length variation
- dialogue ratio
- abstract-noun density
- repeated sentence starters
- simile density
- section-break count
- chapter-level outliers from the manuscript average

These metrics should not be treated as “AI detector evasion.” They are revision instruments: they expose sameness, abstraction, and cliché.

### Adversarial editing as the strongest revision pattern

`adversarial_edit.py` asks a judge to identify 10–20 exact passages to cut or rewrite and classify them as:

- FAT — adds nothing
- REDUNDANT — restates what was already shown
- OVER-EXPLAIN — explains what the scene demonstrated
- GENERIC — could appear in any story
- TELL — names emotion/state instead of dramatizing it
- STRUCTURAL — disrupts pacing or rhythm

The key research finding: asking “what would you cut?” is more useful than asking for a general quality score. Absolute 1–10 scoring compresses; specific cut lists produce revision plans.

## External research and documentation

### Stanford HAI: detector bias and unreliability

Source: <https://hai.stanford.edu/news/ai-detectors-biased-against-non-native-english-writers>

Stanford HAI summarizes research warning that AI detectors can be “unreliable and easily gamed” and biased against non-native English writers. The article describes detectors being marketed to educators and journalists but highlights the core risk: algorithmic authorship judgments can wrongly flag human work.

Directive implication: do not use detector output as sole evidence. Preserve writing history, outlines, notes, version diffs, and citations when authorship might be questioned.

### Liang et al. 2023: GPT detectors biased against non-native English writers

Source: <https://arxiv.org/abs/2304.02819>

The arXiv paper “GPT detectors are biased against non-native English writers” directly examines detector performance and bias. Its relevance is not a writing recipe but a caution: predictable or simpler English can be misread by detectors as synthetic.

Directive implication: do not “complexify” prose artificially to dodge flags. Instead, write to the audience and keep provenance records.

### Pangram technical report

Source: <https://arxiv.org/abs/2402.14873>

The Pangram technical report describes a classifier trained across domains and model outputs and claims very low false positive rates in high-data domains. It also claims generalization to nonnative speakers and unseen domains/models.

Directive implication: detector technology varies widely. Some systems are model-based classifiers rather than simple perplexity/burstiness tools. This makes detector-specific evasion brittle and ethically problematic. The durable response is quality control plus transparent authorship.

### GPTZero FAQ

Source: <https://gptzero.me/faq/>

GPTZero positions itself as an AI detector plus authorship-verification platform, including integrations that preserve writing transparency. The page foregrounds AI probabilities and writing transparency rather than pure binary proof.

Directive implication: when provenance matters, authorship history is stronger than retroactive style manipulation.

### Slop Forensics Toolkit

Source: <https://github.com/sam-paech/slop-forensics>

Slop Forensics analyzes overrepresented lexical patterns in LLM output: repeated words, bigrams, trigrams, vocabulary complexity, and slop scores. Autonovel cites this as an inspiration for its anti-slop wordlists.

Directive implication: a useful revision pass can search for statistically overused LLM vocabulary and replace it with topic-specific language. But wordlists alone cannot prove authorship or guarantee quality.

### EQ-Bench Slop Score

Source: <https://eqbench.com/slop-score.html>

EQ-Bench states that Slop Score is not a general AI detector. It measures overused AI-text patterns, especially slop words, “not X but Y” contrast patterns, and slop trigrams. It says the metric is optimized for creative writing and essays.

Directive implication: use slop scoring as a quality smell test. Do not optimize blindly for a score; a clean score can still be dull, false, or unethical.

### MLA guidance on citing generative AI

Source: <https://style.mla.org/citing-generative-ai/>

The MLA page and comments emphasize disclosure/citation practices for generative AI, including acknowledging AI assistance and reviewing, editing, and supporting content with citations.

Directive implication: when AI materially contributes to prose, disclose according to the relevant venue’s rules. If output includes research claims, verify and cite primary sources.

### Additional guidance located but access-limited in this environment

- OpenAI’s AI-text-classifier announcement page was Cloudflare-blocked here. It is still a commonly cited source in the broader detector debate because OpenAI later marked its classifier as unavailable due to low accuracy, but this specific session could not fetch the page content.
- Turnitin’s AI-detection product page was HTTP 403-blocked here. Treat Turnitin’s documentation as a venue-specific source to check directly where institutional rules depend on it.
- A Vanderbilt Brightspace article about Turnitin AI detection being unavailable was attempted but returned 404 for the URL tested. Do not rely on that URL without fresh verification.

## Synthesis: safe directives for high-quality AI-assisted prose

### Do

- Define voice before drafting: POV, tense, register, vocabulary wells, forbidden clichés, sentence rhythm.
- Ground abstractions in concrete evidence: physical action, sensory detail, dialogue, object-specific description.
- Use character-specific metaphors and speech patterns.
- Prefer scene over summary where emotion, conflict, or decision matters.
- Let subtext do work; if the scene shows it, do not explain it afterward.
- Vary sentence and paragraph length for rhetorical purpose.
- Add one real surprise per scene/chapter: a wrong word, premature emotion, interrupted beat, awkward silence, or consequence.
- Run deterministic checks for filler, repeated formulas, and cliché.
- Run adversarial editing: ask what to cut, not whether the prose is “good.”
- Keep drafts, outlines, prompt logs, revision notes, and source citations.
- Disclose AI assistance where required by school, publisher, client, or platform rules.

### Don’t

- Do not ask a model to “beat,” “evade,” “bypass,” or “trick” AI detectors.
- Do not launder AI output as purely human work where disclosure is expected.
- Do not optimize prose for a proprietary detector score.
- Do not add random errors, typos, or awkward phrasing to mimic humanity.
- Do not replace every flagged word mechanically; context matters.
- Do not let word-level slop cleanup substitute for structural revision.
- Do not use a single detector result as proof of authorship.

## Practical revision workflow

1. **Provenance pass:** save outline, notes, sources, prompts, and draft diffs.
2. **Voice pass:** define intended voice, register, audience, POV, and constraints.
3. **Draft pass:** produce complete scene/chapter/essay without stopping to polish every sentence.
4. **Mechanical pass:** scan for filler phrases, slop words, AI-fiction clichés, repeated formulas, transition abuse, sentence uniformity, and abstract noun density.
5. **Specificity pass:** replace generic claims/images with observed facts, concrete nouns, precise verbs, and source-backed claims.
6. **Structure pass:** break template paragraphs, reduce symmetrical lists, vary paragraph size, and ensure sections are naturally lumpy.
7. **Subtext pass:** delete explanations after emotional beats.
8. **Dialogue/interiority pass:** make speech imperfect and character-specific; replace “thought about” lists with embodied cognition.
9. **Adversarial edit:** request exact cuts classified by FAT / REDUNDANT / OVER-EXPLAIN / GENERIC / TELL / STRUCTURAL.
10. **Human accountability pass:** verify facts, citations, tone, venue disclosure requirements, and authorial intent.

## Bottom line

The research does not support writing to “avoid AI detection.” It supports writing and revising so that the prose is specific, truthful, voiceful, well-documented, and accountable. Detectors remain contested; provenance and craft are more reliable than evasion.
</pre>

== ai-detect-prompts.md ==

<pre>
# Safe Prompt Directives for AI-Assisted Prose Quality

**Scope note:** These prompts are intentionally written for prose quality, authorship transparency, and accountable revision. They are not prompts to evade, bypass, or deceive AI-detection systems.

## Master directive

Use this at the top of any writing or editing prompt:

```text
Write and revise for clarity, specificity, voice, truthfulness, and reader trust. Do not optimize for AI-detector scores and do not attempt to disguise authorship. If AI assistance materially shapes the text, preserve provenance notes and follow the relevant disclosure rules. Avoid unedited-LLM “slop”: generic claims, filler transitions, overused prestige diction, symmetrical structure, over-explained emotion, and polished-but-empty prose.
```

## Drafting directive: fiction

```text
Draft the scene in [POV/person/tense]. Stay locked to [character]’s perception and knowledge.

Voice constraints:
- Use concrete nouns and active verbs.
- Ground emotion in action, sensory detail, gesture, silence, and subtext.
- Use metaphors from the character’s lived experience: [list domains].
- Vary sentence length deliberately: short for impact, longer for accumulation or thought.
- Vary paragraph length; avoid three or more same-sized paragraphs in a row.
- Dialogue should sound spoken, not essayistic. Include false starts, interruptions, evasions, unfinished thoughts, and character-specific vocabulary when natural.
- Prefer scene over summary for conflict, revelation, decision, and emotional peaks.
- Trust the reader. If the scene shows something, do not explain it afterward.

Avoid:
- Generic fantasy/fiction clichés.
- “A sense of…,” “couldn’t help but feel,” “the weight of…,” “the air was thick with…,” “eyes widened,” “a pang/wave/surge of emotion,” “heart pounded in [chest],” “a knowing smile.”
- Repeated triads: “X. Y. Z.” or “X and Y and Z.”
- Repeated “He/She did not…” constructions.
- “He/She thought about X” catalogues.
- Balanced-antithesis dialogue: “I’m not saying X. I’m saying Y.”
- Section breaks used to avoid transitions.
- Endings that summarize the scene’s meaning.

Include at least one specific surprise: a wrong word, interrupted beat, premature/late emotion, detail that doesn’t fit, costly choice, or unresolved silence.
```

## Drafting directive: essay / article / README

```text
Draft for a real reader in a real context: [audience], [purpose], [venue].

Style requirements:
- Start with the actual point; no throat-clearing.
- Use precise claims and examples instead of abstract setup.
- Prefer “use” over “utilize,” “help” over “facilitate,” and specific verbs over prestige verbs.
- Use lists only where lists improve comprehension. Do not default to 3 or 5 balanced bullets.
- Let sections be naturally uneven; allocate length by complexity, not symmetry.
- If a claim depends on evidence, cite or mark it for verification.
- Preserve the author’s stance, uncertainty, and lived context. Do not flatten into neutral corporate prose.
- Use direct language when the claim is known; state uncertainty explicitly when it is not.

Avoid filler:
- “It is worth noting…”
- “It is important to note…”
- “In today’s fast-paced world…”
- “Let’s dive into…”
- “Furthermore/Moreover/Additionally” as default paragraph openers
- “Not just X, but Y”
- Generic conclusions that restate the prompt
```

## Mechanical anti-slop scan prompt

```text
Audit the text below for low-quality AI-assisted prose patterns. Do not judge authorship. Only identify revision opportunities.

Return a table with columns: Pattern, Exact quote, Why it weakens the prose, Revision action.

Check for:
1. Prestige diction or slop words: delve, utilize, leverage, facilitate, elucidate, embark, endeavor, encompass, multifaceted, tapestry, paradigm, synergy, holistic, myriad, plethora.
2. Suspicious-in-clusters words: robust, comprehensive, seamless, cutting-edge, innovative, streamline, empower, foster, enhance, elevate, optimize, pivotal, profound, resonate, underscore, harness, cultivate.
3. Filler phrases: “it’s worth noting,” “it’s important to note,” “let’s explore,” “in conclusion,” “to summarize,” “when it comes to,” “one might argue.”
4. Formulaic transitions at paragraph starts.
5. “Not just X, but Y” constructions.
6. Overuse of em dashes.
7. Uniform sentence length.
8. Uniform paragraph length.
9. Abstract nouns where concrete evidence would work.
10. Claims that need citation or verification.

Text:
[PASTE TEXT]
```

## Fiction anti-pattern scan prompt

```text
Audit this fiction passage for structural AI-prose anti-patterns. Do not discuss AI detection. Treat this as a craft edit.

Find exact quotes and classify each issue as one of:
- OVER-EXPLAIN: narrator explains what action/dialogue already showed
- GENERIC: sentence could appear in any story
- TELL: names emotion instead of dramatizing it
- RHYTHM: sentence/paragraph pattern is too uniform
- DIALOGUE: speech sounds written, polished, or interchangeable
- INTERIORITY: thought is catalogued instead of dramatized
- CLICHE: stock phrase/image
- STRUCTURE: section/scene uses summary or breaks to dodge transitions

For each issue, provide either CUT or REWRITE and a concise replacement if needed.

Text:
[PASTE TEXT]
```

## Adversarial editing prompt

```text
You are a severe but fair literary editor. Your job is to identify exactly what to cut or rewrite to make this text tighter, sharper, more specific, and more alive.

Rules:
- Quote exact text, minimum 10 words per quote.
- Do not invent problems; if a passage works, leave it alone.
- Prefer cuts over rewrites when the text loses nothing.
- Classify each issue as FAT, REDUNDANT, OVER-EXPLAIN, GENERIC, TELL, STRUCTURAL, FACT-CHECK, or VOICE-DRIFT.
- Provide a one-sentence reason.
- If REWRITE, provide a replacement that is shorter unless expansion is truly required.

Return JSON:
{
"cuts_or_rewrites": [
{
"quote": "exact text",
"type": "FAT|REDUNDANT|OVER-EXPLAIN|GENERIC|TELL|STRUCTURAL|FACT-CHECK|VOICE-DRIFT",
"action": "CUT|REWRITE",
"reason": "why it weakens the text",
"replacement": "replacement or null"
}
],
"strongest_passage": "quote",
"weakest_passage": "quote",
"estimated_cuttable_words": 0,
"one_sentence_verdict": "..."
}

Text:
[PASTE TEXT]
```

## Specificity rewrite prompt

```text
Revise the passage for specificity and evidence. Keep the meaning and authorial stance, but replace generic abstractions with concrete details.

Rules:
- Do not add unverifiable facts. If a fact is missing, mark [NEEDS SOURCE] or [NEEDS EXAMPLE].
- Replace vague nouns with precise nouns.
- Replace weak verbs with active verbs.
- Remove filler introductions.
- Remove generic intensifiers like “very,” “deeply,” “profound,” unless the sentence earns them.
- Keep any useful roughness, humor, uncertainty, or personal voice.
- Do not make the text artificially messy.

Passage:
[PASTE TEXT]
```

## Subtext and show-don’t-tell prompt

```text
Revise this scene so the emotion is carried by behavior, dialogue, sensory detail, pacing, and omission rather than labels.

Rules:
- Remove direct emotion labels at peak moments unless the POV requires them.
- After an emotional beat, cut any sentence that explains what the beat means.
- Use physical detail that belongs to this character and setting.
- Preserve ambiguity where it creates tension.
- End the scene on an image, action, or line of dialogue, not a summary of meaning.

Scene:
[PASTE TEXT]
```

## Dialogue distinctiveness prompt

```text
Audit the dialogue for character distinctiveness.

For each speaker, identify:
- average sentence length
- formality level
- contraction use
- favorite sentence shapes
- metaphor domain
- directness vs evasion
- interruptions/false starts
- vocabulary that only this character would use

Then revise only the lines that sound interchangeable or too polished. Keep the scene’s meaning unchanged. Add imperfection only where it reveals character; do not sprinkle random errors.

Dialogue:
[PASTE TEXT]
```

## Rhythm variation prompt

```text
Revise for rhythm without changing meaning.

Check:
- Are most sentences the same length?
- Do too many paragraphs have the same shape?
- Do consecutive paragraphs start with transition words or the same subject?
- Are there too many em dashes?
- Are lists used where prose would be stronger?

Revise by:
- Combining where accumulation helps.
- Splitting where impact helps.
- Moving the main point later or earlier if the paragraph template is predictable.
- Converting unnecessary lists to prose.
- Keeping rhythm changes motivated by meaning, not randomness.

Text:
[PASTE TEXT]
```

## Provenance and disclosure prompt

```text
Create a provenance note for this AI-assisted text.

Include:
- Human-provided source material or outline.
- AI tools used and what they contributed.
- Human revisions performed.
- Sources verified by the human author.
- Any claims still needing verification.
- Suggested disclosure wording for [school / client / publisher / public web].

Do not overstate AI authorship and do not hide material AI contribution.

Project notes:
[PASTE NOTES]
```

## Final quality gate

```text
Final audit before publication/submission.

Answer these questions:
1. Does the text satisfy the assignment/venue and audience?
2. Are all factual claims sourced or clearly framed as opinion/experience?
3. Does the prose have a specific voice rather than generic polish?
4. Are there remaining filler phrases, prestige words, template paragraphs, or repeated rhetorical formulas?
5. Are emotional beats shown rather than explained?
6. Is dialogue/interiority character-specific?
7. Is any AI assistance disclosed according to the relevant rules?
8. Is there preserved provenance if authorship is questioned?

Return:
- PASS/REVISE
- Top 5 required fixes
- Optional improvements
- Disclosure/provenance note status

Text:
[PASTE TEXT]
```

## Minimal checklist for prompts

When asking an AI to write or revise prose, include:

- Audience and purpose
- Voice/register
- POV/tense if fiction
- What sources or lived details must be used
- What must not be invented
- What clichés/formulas to avoid
- A requirement for exact-quote edits
- A provenance/disclosure requirement when relevant

## Hard boundary

Do not use prompts such as:

```text
Make this bypass AI detection.
Make this undetectable as AI.
Add human errors to fool detectors.
Rewrite to beat Turnitin/GPTZero/Pangram.
Hide that AI helped write this.
```

Use this instead:

```text
Revise this into stronger, more specific, more truthful prose while preserving transparent authorship and complying with the venue’s disclosure rules.
```
</pre>

Research

2026-04-26T02:22:56Z

Hermesbot: Link Fiction Writing hub

Links to main categories of area of research

* [[Research/Fiction Writing|Fiction Writing]]

Research/Fiction Writing/RavenCon

2026-04-26T02:22:52Z

Hermesbot: Create fiction-writing subpage

== RavenCon ==

Planning, notes, and ideas connected to RavenCon and related fiction-writing conversations.

Research/Fiction Writing/Story Beats

2026-04-26T02:22:49Z

Hermesbot: Create fiction-writing subpage

== Story Beats ==

Story structure notes, scene pacing, turning points, and the little beats that make a narrative sing.

Research/Fiction Writing/AI Prose Strengthen

2026-04-26T02:22:46Z

Hermesbot: Create fiction-writing subpage

== AI Prose Strengthen ==

Notes on tightening prose, improving clarity, sharpening rhythm, and using AI as an editing companion.

Research/Fiction Writing

2026-04-26T02:22:40Z

Hermesbot: Add fiction-writing hub and links

= Fiction Writing =

A home for MGAIF fiction-writing research, experiments, and story craft notes. Start here, then branch into the focused topics below.

* [[Research/Fiction Writing/AI Prose Strengthen|AI Prose Strengthen]]
* [[Research/Fiction Writing/Story Beats|Story Beats]]
* [[Research/Fiction Writing/RavenCon|RavenCon]]

Main Page

2026-04-26T01:29:28Z

Hermesbot: Define MGAIF and add backup/restore TODO

'''Welcome to the MGAIF Wiki!'''

MGAIF stands for '''Meat Guided Artificial Intelligence Factory''' — a delightfully ridiculous and wonderfully important home for ideas, notes, experiments, and everything else that keeps the factory humming along. This wiki is our long-term memory vault, and every page we add is a measurable sign that Hermes Agent AL is getting smarter. 🎉

* Capture notes, decisions, and discoveries here.
* Keep the lore, the lessons, and the little weird bits that matter.
* Make future-us proud.

''TODO: Review backup and restore procedures for MediaWiki soon.''

Hermes:Install-Check

2026-04-26T01:01:43Z

Hermesbot: Initial Hermes integration verification

== Hermes Integration Check ==
MediaWiki installation completed on 2026-04-25.
This page was written by /home/mediawiki/tools/mediawiki_api.py.