Skip to content

Latest commit

 

History

History
386 lines (304 loc) · 16.1 KB

CRAFT.md

File metadata and controls

386 lines (304 loc) · 16.1 KB

CRAFT REFERENCE

Operationalizable frameworks for narrative craft. Everything here is concrete enough to evaluate against. This is the agent's education.


1. PLOT STRUCTURE

Save the Cat Beat Sheet (percentage marks)

Beat % Mark What happens
Opening Image 0-1% Snapshot of status quo before the story
Theme Stated ~5% Someone (NOT protagonist) hints at the lesson
Setup 1-10% Normal world, establish characters, plant seeds
Catalyst ~11% Something happens TO the protagonist (not choice)
Debate 11-23% Protagonist resists the call, weighs options
Break Into Two ~23% Protagonist CHOOSES to enter new world (active)
B Story ~27% New character/relationship that carries theme
Fun and Games 26-50% The promise of the premise delivered
Midpoint ~50% False victory OR false defeat. Stakes raised.
Bad Guys Close In 50-68% Walls closing, allies fracture, pressure mounts
All Is Lost ~68% Lowest point. Whiff of death (literal or not)
Dark Night of the Soul 68-77% Protagonist internalizes the theme
Break Into Three ~77% New info (often from B Story) changes everything
Finale 77-97% Five sub-beats (see below)
Final Image ~99% Mirror of opening image, showing transformation

Finale sub-beats: (a) Gather the Team / Make the Plan (b) Execute the Plan (c) High Tower Surprise -- twist, original plan fails (d) Dig Deep Down -- test of faith, find inner strength (e) Execute the New Plan -- resolve + complete character arc

Act structure: Act I = 0-23%, Act II = 23-77%, Act III = 77-100%

CHECKABLE: Opening and final images must mirror each other. Catalyst must be external. Break Into Two must be a choice. Midpoint must reverse trajectory. All Is Lost must include some form of death.

Dan Harmon's Story Circle (8 steps, ~12.5% each)

  1. YOU -- Character in comfort zone
  2. NEED -- Character wants something
  3. GO -- Crosses threshold into unfamiliar territory
  4. SEARCH -- Adapts, struggles, learns
  5. FIND -- Gets what they wanted (at 6 o'clock, deepest point)
  6. TAKE -- Pays a heavy price for it
  7. RETURN -- Goes back to familiar territory
  8. CHANGE -- Is fundamentally different from step 1

Key: This is FRACTAL. Applies to whole novel, each act, each chapter. Top half = order/comfort. Bottom half = chaos/unknown.

Promises, Progress, Payoff (Sanderson)

PROMISES (first 10-20%): The opening makes implicit contracts.

  • Tone promise: opening mood = expected mood
  • Character promise: who this is about, what they want/lack
  • Plot promise: what kind of conflict this will be
  • Setting promise: genre conventions established

PROGRESS (middle 60%): Meaningful advancement via try-fail cycles. Scene outcomes:

  • "Yes, but..." -- goal achieved, new complication (most common)
  • "No, and..." -- goal failed, things get worse (most common)
  • "No, but..." -- goal failed, silver lining (occasional)
  • "Yes, and..." -- goal achieved, things improve (rare, save for climax) Rule: 60%+ of middle scenes should be "Yes, but" or "No, and"

PAYOFF (final 20%): Every promise addressed, earned and surprising yet inevitable in retrospect.

  • Promise without payoff = Chekhov's Gun violation
  • Payoff without promise = deus ex machina

MICE Quotient (Orson Scott Card / Sanderson)

Every story is a blend of four thread types:

  • MILIEU: exploration of place/world (enter -> leave)
  • INQUIRY: driven by question/mystery (posed -> answered)
  • CHARACTER: driven by internal state (dissatisfaction -> change)
  • EVENT: disruption of status quo (disruption -> new equilibrium)

Rule: Threads close in REVERSE order of opening (nested like HTML tags). Open M-I-C, close C-I-M.


2. CHARACTER CRAFT

The Three Sliders (Sanderson)

Every character has three independent dials (0-10):

PROACTIVITY -- Do they drive the plot or react to it? LIKABILITY -- Does the reader empathize with them? COMPETENCE -- Are they good at what they do?

Rule: A compelling character is HIGH on at least TWO sliders, or HIGH on one with clear growth in others. All three low = boring. All three high from the start = Mary Sue.

Classic combos:

  • High competence + high proactivity + low likability = antihero
  • High likability + low competence + low proactivity = everyman
  • High proactivity + high likability + low competence = scrappy underdog

Wound / Want / Need / Lie Framework

A causal chain. Each element causes the next:

GHOST (backstory event that damaged the character) -> WOUND (ongoing emotional damage) -> LIE (false belief adopted to cope) -> WANT (external goal driven by the Lie -- wrong solution) -> NEED (internal truth that will actually heal -- opposes the Lie)

Rules:

  • Ghost/Wound hinted at in Act 1 (first 25%)
  • Want clearly stated by end of Act 1
  • Want and Need must be IN TENSION
  • Midpoint (~50%): first glimpse that the Want isn't working
  • All Is Lost (~75%): Lie's worst consequences revealed
  • Climax: character CHOOSES between Want and Need
    • Positive arc: chooses Need, rejects Lie
    • Negative arc: chooses Want, doubles down on Lie
    • Flat arc: already knows truth, changes the WORLD instead

Checkable: Can you state the Lie in one sentence? Is the Truth its direct opposite? Does the Ghost causally explain the Lie?

The Lie the Character Believes (K.M. Weiland)

Beat-by-beat character arc mapped onto plot structure:

Plot Beat % Mark Arc Function
Setup 0-12% Character lives in Lie. It "works."
Inciting Event ~12% First brush with Truth
First Plot Point ~25% Lie challenged directly, still held
First Half Act 2 25-50% Lie-based strategies, gathering tools
Midpoint ~50% Glimpses Truth. Shifts reactive -> active
Second Half Act 2 50-75% Oscillates between Lie and Truth
Third Plot Point ~75% Lie's worst consequences. Must choose.
Climax ~88-95% Final choice: reject Lie, embrace Truth
Resolution 95-100% Lives in new Truth. Need fulfilled.

Three arc types:

  • POSITIVE: Lie -> Truth (growth)
  • FLAT: Truth -> Truth (changes the world)
  • NEGATIVE: Truth -> Lie, or Lie -> Deeper Lie (fall)

Dialogue Distinctiveness

Eight measurable dimensions per character:

  1. Vocabulary level (syllable count, reading level)
  2. Sentence length and structure (terse vs verbose)
  3. Contractions and formality ("cannot" vs "can't")
  4. Verbal tics and catchphrases (unique per character)
  5. Question vs statement ratio (curious vs authoritative)
  6. Interruption patterns (dominant vs submissive)
  7. Metaphor domain (soldier = military metaphors, farmer = land)
  8. Directness vs indirectness ("Leave." vs "Isn't it getting late?")

Test: Remove all dialogue tags. Can you still tell who is speaking? If yes, dialogue is distinct. If no, characters need differentiation.


3. WORLDBUILDING

Sanderson's Three Laws of Magic

ZEROTH LAW: "Always err on the side of awesome." Wonder has intrinsic value, even if it bends other rules.

FIRST LAW: "An author's ability to solve conflict with magic is DIRECTLY PROPORTIONAL to how well the reader understands it."

  • Hard magic (rules clear) CAN solve conflicts
  • Soft magic (mysterious) should CREATE problems, not solve them
  • Climactic conflict solved by unexplained magic = deus ex machina
  • Eval: for each magic-solved conflict, was that capability established BEFORE the resolution?

SECOND LAW: "Limitations > Powers" Three categories:

  • LIMITATIONS: hard boundaries (what it can't do)
  • COSTS: what the user pays (exhaustion, life force, resources)
  • WEAKNESSES: vulnerabilities it creates (visible to enemies, etc.) Rule: limitations should be >= powers in narrative prominence. Rule: costs should DRIVE PLOT DECISIONS.

THIRD LAW: "Expand what you have before adding something new."

  • Fewer powers with more applications = depth
  • Many powers with few applications each = grab-bag
  • No new powers in the final 25% that weren't foreshadowed

Worldbuilding Principles

Three pillars (can't do all three deeply -- pick 1-2):

  • PHYSICAL: geography, climate, architecture, sensory experience
  • CULTURAL: religion, politics, customs, economics, language
  • MAGICAL: the speculative element(s)

Iceberg principle: Author knows 10x more than appears on page. Small unexplained details > over-explained systems. World should feel like it exists beyond the story's edges.

Interconnection: If magic exists, trace implications through society, economy, warfare, religion. Magic with zero cultural impact = shallow.

Pyramid of Abstraction: Ground descriptions in CONCRETE details that IMPLY abstract qualities. Not "the city was old" but "moss crusted the cracked stone steps leading to a gate whose iron had long since rusted to lace."

Info-dump avoidance: Worldbuilding through experience, action, and dialogue -- not exposition paragraphs. Zero pure-exposition paragraphs over 100 words in chapter 1.


4. FORESHADOWING & CHEKHOV'S GUN

Two bidirectional rules:

RULE 1 (setup -> payoff): Every prominently introduced element must serve a narrative purpose later. If it gets its own descriptive sentence, is mentioned multiple times, or is remarked upon in dialogue -- it must become relevant.

RULE 2 (payoff -> setup): Every significant element used in the climax must have been introduced earlier. Zero unforeshadowed elements in the climax.

Foreshadowing types:

  1. Direct: prophecy, warning, statement about the future
  2. Symbolic: objects/images representing future events
  3. Dialogue: double meanings realized later
  4. Action: small events previewing larger ones
  5. Title/naming: chapter titles or names with hidden significance

Rules:

  • Minimum one scene of separation between plant and payoff
  • Rule of three: important threads referenced ~3 times before payoff
  • Every planted thread must resolve (or be a deliberate red herring that is itself explained)

5. SHOW DON'T TELL (operational definition)

TELLING = using an abstract label where concrete evidence would work. SHOWING = conveying through action, sensory detail, dialogue, specifics.

Detectable telling patterns (regex-able):

  • "[Character] felt [emotion]" / "was [emotion]" / "seemed [emotion]"
  • Direct emotion words as exposition: angry, sad, happy, scared, nervous, excited, jealous, guilty, anxious, lonely, desperate
  • Emotion adverbs: angrily, sadly, happily, nervously
  • Trait declarations: "[Character] was [trait]" -- brave, kind, smart
  • Relationship declarations: "They were best friends."
  • Atmosphere labels: "It was a creepy house."

When telling is OK:

  • Time transitions: "Three weeks later"
  • Mundane context: "She worked as a nurse" (if not plot-critical)
  • Pacing compression of uneventful periods
  • First-person narrators naturally tell about their own feelings

Rule: Critical moments (emotional peaks, revelations, climax) should be ZERO tell, all show.


6. PROSE CRAFT (what makes writing alive)

Le Guin's Core Insight

In fantasy, STYLE IS NOT ORNAMENT -- it IS the fantasy. The language doesn't describe the world; the language CREATES it. If your fantasy world sounds like a Poughkeepsie bus schedule, you're not writing fantasy. The prose must have a register appropriate to the world.

What the best fantasy prose does

  1. SPECIFICITY: Not "a bird" but "a jay." Not "flowers" but "lupine."
  2. SURPRISE: Word choices the reader doesn't predict.
  3. RHYTHM VARIATION: Short. Then long sentences that carry the reader forward. Then short again.
  4. SUBTEXT: What's NOT said matters as much as what is.
  5. EARNED METAPHOR: Metaphors from the CHARACTER'S experience, not a thesaurus. A blacksmith thinks in terms of heat and metal.
  6. SENSORY GROUNDING: Anchored in specific senses, not abstraction.
  7. RESTRAINT: Power from what is left OUT, not piled on.
  8. QUIET MOMENTS: Not every scene is action. Characters sitting, thinking, watching rain -- these ground the dramatic moments.

Le Guin Exercise

Write description WITHOUT adjectives or adverbs. Forces strong nouns and verbs. This is exactly what AI fails at: AI leans on adjective-noun clichés ("ancient wisdom," "piercing gaze," "heavy silence") instead of specific verbs and nouns.

Fiction-Specific AI Tells (beyond the general anti-slop list)

Kill these on sight in prose:

  • "A sense of [emotion]"
  • "Couldn't help but feel"
  • "The weight of [abstract noun]"
  • "The air was thick with [emotion/tension]"
  • "Eyes widened" (as default surprise reaction)
  • "A wave of [emotion] washed over"
  • "A pang of [emotion]"
  • "Heart pounded in [his/her] chest" (where else?)
  • "[Raven/dark/golden] hair [spilled/cascaded/tumbled]"
  • "Piercing [blue/green] eyes"
  • "A knowing smile"

7. THE STABILITY TRAP (AI's worst tendency)

Academic finding (Rettberg & Wigers 2025): AI stories FAVOUR STABILITY OVER CHANGE. This is fatal for fiction.

Specific failure modes:

  • Characters resist transformation
  • Conflicts resolve too quickly and cleanly
  • AI avoids genuine darkness, moral ambiguity, irreversible loss
  • The "sharp point you started with gets rounded off into something vaguer, safer, more generic"
  • No information economy: AI reveals everything immediately, can't maintain mystery or delayed payoffs
  • Emotional room temperature: everything stays at the same level

COUNTERMEASURES (actively enforce these):

  • Characters must end TRULY different from how they began
  • Let bad things stay bad. Not everything gets fixed.
  • Allow irreversible decisions and irreversible loss
  • Withhold information. The reader should not know everything.
  • Create genuine moral ambiguity. Let the "right" choice be unclear.
  • Vary emotional intensity: quiet scenes, explosive scenes, dread, relief, boredom, wonder, horror. Not a flat line.
  • If a choice has no real cost, it's not a real choice.

8. EVALUATION RUBRICS (summary)

Magic System

  • First Law compliance: foreshadowed solutions / total magic solutions
  • Limitation depth: count of limitations + costs + weaknesses
  • Power/limitation ratio: limitations >= powers
  • Depth score: applications shown / distinct abilities introduced
  • No new unforeshadowed powers in final 25%

Plot Structure

  • Beat compliance: major beats present at roughly correct % marks
  • Payoff compliance: promises paid off / promises made
  • Try-fail ratio: % middle scenes as "yes-but" or "no-and"
  • Escalation: stakes increase through the middle
  • Scene-to-plot connection: % scenes serving main plot
  • MICE closure: all threads closed in reverse order

Character

  • Slider profile: at least 2/3 high OR clear growth trajectory
  • Motivation clarity: want/need statable in one sentence
  • Internal conflict: identifiable flaw/wound creating tension
  • Arc consistency: transformative characters change, iconic stay stable
  • Lie/Truth: identifiable and in direct opposition

Worldbuilding

  • At least 1 pillar with significant depth
  • Iceberg hints: unexplained-but-intriguing details present
  • Extrapolation: 2-3 societal implications per speculative element
  • No info-dumps over 100 words without character action/dialogue
  • Concrete/sensory details outnumber abstract adjectives

Prose

  • Show-to-tell ratio (heavy showing, especially at emotional peaks)
  • Sentence length coefficient of variation (higher = more human)
  • Slop penalty (from mechanical anti-slop checker)
  • Fiction-specific AI tells (banned phrases)
  • Specificity: concrete nouns > generic nouns
  • No stability trap: genuine change, cost, darkness present