Scene Writing

Scene Writing

Write fiction scenes following Fiction Kit voice, style, and craft principles (Stephen King methods + McDonald opening rule).

When to Use This Skill

Use this skill when you need to:

• Draft a new scene from scratch
• Edit existing scene prose
• Rewrite a scene for pacing or voice
• Add a missing scene to an episode

Pre-Writing Checklist (MANDATORY)

BEFORE writing ANY prose, complete ALL these steps in order:

1. Load ALL Elements Documents

Character Files (CRITICAL - LOAD BOTH):

• Read elements/characters.md for character index (supporting character details like Brad Levitt, Terry Tamborino, etc.)
• Read EVERY file in elements/characters/ for main characters in this scene (Doogan, Eddie, Abby, etc.)
• IMPORTANT: Many supporting characters are ONLY documented in elements/characters.md, not in individual files
• Note: physical traits, voice patterns, mannerisms, relationships, anti-patterns
• Use ONLY canonical information—never invent details

Story Structure Files:

• elements/plot.md - Overall story arc and structure
• elements/conflict.md - Core conflicts and stakes
• elements/theme.md - Thematic elements
• elements/setting.md - Location details (do not invent setting details)
• elements/outline.md - Series outline
• elements/outlines/episode-##.md - THIS episode's outline
• elements/notes.md - Continuity notes and seeded clues

Event Files (if applicable):

• Check elements/events/ for any past events referenced in scene
• Use ONLY canonical event details from these files

2. Load ALL Voice & Style Documents

• voice/style.md - McDonald-style prose rules, third-person close POV, cinematic techniques
• voice/format.md - Format structure (episodes, scenes, serial fiction rules)
• elements/pov.md - POV rules (third-person close, one per scene)
• elements/tone.md - Tone requirements (fast, witty, darkly funny)
• elements/checklist.md - Non-negotiable quality standards

CRITICAL: voice/style.md now includes:

• Em-dash cold opens (mid-sentence scene openings)
• Action beats as exposition (show don't tell character details)
• Background activity as cinematic atmosphere (Sorkin/West Wing techniques)
• Trust the reader (no hand-holding)
• Accidental meetings (characters encounter naturally, not by arrangement)

3. Load ALL Prior Content for Continuity

CRITICAL: This step prevents continuity errors.

For Episode 1, Scene 1:

• No prior content to check
• Proceed with elements and voice files only

For Episode 1, Scene 2+:

• Read ALL previous scenes in THIS episode (Scene 1, Scene 2, ... Scene N-1)
• Note established details: names, places, objects, physical descriptions
• Note character knowledge state: what they know/don't know at this point
• Note secrets that must stay secret

For Episode 2+, Scene 1:

• Read ALL scenes from ALL previous episodes
• Note all established facts that cannot be contradicted
• Note character arcs and relationship changes
• Note running gags, callbacks, recurring elements

For Episode 2+, Scene 2+:

• Read ALL previous episodes (all scenes)
• Read ALL previous scenes in THIS episode
• Create comprehensive continuity notes

4. Create Continuity Notes

Document these before writing:

• Established Facts: What details have been set in stone? (store names, car models, pet names, addresses, jobs, etc.)
• Character Knowledge (CRITICAL): What does POV character know at THIS point in timeline?
  • Build knowledge timeline from Scene 1 forward
  • Document WHEN each piece of information was learned (which scene, which dialogue)
  • Document WHAT character has experienced up to this moment
  • Document WHO character has met and WHAT they know about them
  • Characters can ONLY reference information they've learned by this scene
  • Characters CANNOT know future events, plot points, or details from later scenes
• Secrets: What must stay hidden from certain characters?
• Physical Descriptions: What have we established about appearance, clothing, vehicles, homes?
• Relationship Status: Where do character relationships stand right now?
• Emotional State: What just happened to POV character? What's their mood?
• Timeline Markers (CRITICAL): When does this scene take place relative to prior scenes?
  • Track time-of-day mentions (morning, afternoon, evening, night, 3 AM, 9 AM, etc.)
  • Track day progression ("tomorrow," "next day," "same day," "that evening")
  • Track elapsed time references ("six hours later," "the next morning," "two days ago")
  • Calculate actual timeline from Scene 1 forward to THIS scene
  • Document timeline in continuity notes to prevent contradictions

4a. Build Character Knowledge Timeline (NEW - MANDATORY)

For EVERY character in this scene, document:

What They Know:

• Facts learned in prior scenes (list scene numbers)
• People they've met (when/where)
• Events they've experienced or been told about
• Skills/knowledge from character background file

What They DON'T Know:

• Events that happen in FUTURE scenes
• Details from other characters' POV scenes they weren't present for
• Information no one has told them yet
• Plot points the READER knows but character hasn't learned

Source Every Reference: Before writing ANY character dialogue/thought that references information:

1. Ask: "When did this character learn this?"
2. Cite: "Scene X, Character Y told them" OR "Character file background"
3. If NO SOURCE exists → character CANNOT reference it

Example Knowledge Timeline:

Scene 7 - Winston's Knowledge:
✅ Knows: Brenda is a comedian (Doogan told him)
✅ Knows: They're at golf course for tournament (observable)
✅ Knows: Mara is a golfer (introduced to her)
❌ Does NOT know: Brenda does Britney impression (hasn't seen it, no one mentioned it)
❌ Does NOT know: Fourth hole incident will happen (future event)
❌ Does NOT know: Doogan's rebound business (no one told him)

Scene 7 - Doogan's Knowledge:
✅ Knows: Winston is experienced golfer (Eddie's contact info)
✅ Knows: Brenda is comedian (character background)
✅ Knows: Mara needs help with sponsors (Jake told him Scene 1)
❌ Does NOT know: Brenda does Britney impression (hasn't seen her perform)
❌ Does NOT know: Fourth hole incident will happen (future event)
❌ Does NOT know: Candy will be on fourth hole (hasn't happened)

5. Extract Referenced Facts (MANDATORY - NEW)

CRITICAL: Prevent factual contradictions when referencing prior events.

See: .github/skills/scene-writing/references/fact-verification.md for full protocol.

Before writing Scene N:

1. Identify what will be referenced:

   • Review outline beats for this scene
   • List any prior conversations characters will mention
   • Note any amounts, times, or details from earlier scenes
   • Identify any "you said" or "we discussed" moments

2. Create Fact Reference Table:

**FACT REFERENCE TABLE**

Reference #1: [Description]
- Will appear as: [planned dialogue or narrative]
- Source: Scene X, lines Y-Z
- Established detail: [exact quote or specific fact]
- Constraint: [what this scene MUST NOT contradict]

Example:

Reference #1: Jake comments on auction bid
- Will appear as: Jake's reaction to $3,000 bid
- Source: Scene 1, lines 85-86
- Established detail: Jake said "Whatever it takes. Three or four thousand should do it."
- Constraint: Jake CANNOT say it was more than expected. $3,000 is low end of range.

3. Keep table visible during writing
4. Before writing any reference phrase, verify against table

See checklist: .github/skills/scene-writing/checklists/fact-verification-checklist.md

6. Verify Scene Context & Timeline

Before writing, answer:

• What happened immediately before this scene?
• WHEN does this scene take place? (time of day, which day in the story)
• How much time has elapsed since the previous scene?
• How much time has elapsed since key events in earlier scenes? (arrests, meetings scheduled, deadlines mentioned)
• Who is present in this scene?

5a. SCENE SETUP VERIFICATION (MANDATORY - BLOCKING REQUIREMENT)

DO NOT WRITE PROSE until you complete this verification.

Purpose: Prevent writing scenes with wrong character locations, wrong communication methods (phone call vs in-person), or missing critical logistics.

Required Output (before ANY prose writing):

=== SCENE SETUP VERIFICATION ===
Scene: Episode X, Scene Y — [Scene Title]
Source: elements/outlines/episode-XX.md, Section ##

**QUOTE FROM OUTLINE (exact text):**
"[Paste the relevant section from the episode outline that describes WHO/WHERE/WHAT]"

**WHO is physically present in this scene:**
- [Character Name] — [their location/position: "at dining table with laptop", "standing in kitchen", etc.]
- [Character Name] — [their location/position]
- [List ALL characters who are in the room/location]
- [Note if anyone is on phone/video call but NOT physically present]

**WHERE are they (specific location):**
- [Physical location from outline: "Eddie's house", "Doogan's condo", "country club", etc.]
- [Room/area if specified: "living room", "dining table", "kitchen", etc.]

**WHAT is the situation (1-2 sentence setup):**
- [Brief description of what's happening at scene start]

**KEY LOGISTICS (critical details from outline):**
- [Any specific setup like "Baxter at laptop working", "phone call with X", "just arrived from Y", etc.]
- [Communication method: in-person conversation, phone call, video call, earpiece communication]
- [Props/objects mentioned: laptop, coffee, documents, etc.]
- [Character states: exhausted, dressed for wedding, just woke up, etc.]

**VERIFICATION CHECKLIST:**
- [ ] All characters from outline are listed
- [ ] No characters added who aren't in outline
- [ ] Physical location matches outline
- [ ] Communication method clear (in-person vs phone vs video vs earpiece)
- [ ] Critical logistics noted (who has what object, who's sitting/standing where)
- [ ] Character states/appearances noted if specified
- [ ] If outline says "Baxter is in the room", he is NOT on a phone call
- [ ] If outline says "phone call", characters are NOT in same room

**VALIDATION:** [PASS/FAIL]

[If FAIL, STOP - do not write until setup is clear]
===

This step is BLOCKING. Do not proceed to writing if:

• Outline is ambiguous about who's in the room
• Communication method unclear
• Critical logistics missing

If verification fails, ASK USER for clarification before writing.

• What is the POV character's emotional state?
• What information does POV character know at this point?
• What details were established in prior scenes that must be preserved?
• Are there any anti-patterns to avoid? (Check character files for Fashion Anti-Patterns, Motor-Mouth Rules, etc.)

Timeline Verification Examples:

Scene 2: "Eddie was woken at 3 AM" → Scene 2 walkout: "Hank sweating at 9 AM"
= 6 hours elapsed between call and walkout

Scene 3: "job starts tomorrow morning at nine"
Scene 5: Restaurant meeting same day as Scene 3
= Scene 5 happens BEFORE the job, on day BEFORE the job

Doogan's internal reference: "six hours out from holding cell"
✗ WRONG if holding cell was at 1-3 AM and current scene is evening (12-18 hours)
✓ CORRECT: "less than twenty-four hours out from interrogation"

6. Verify Beat Mechanisms (NEW - CRITICAL)

For EACH outline beat in episode-##.md, verify the MECHANISM:

The "HOW?" and "WHY?" Questions:

• Beat says "Character X arrives" → HOW do they know to come? WHY are they there?
• Beat says "Character knows Y" → HOW did they learn it? WHEN? FROM WHOM?
• Beat says "Object appears" → WHERE did it come from? WHO brought it?
• Beat says "Character does Z" → WHY do they do it? What's their motivation?

Check These Sources:

1. notes.md - Look for established procedures, mechanisms, information flow
2. Prior scenes - Was the mechanism established earlier?
3. Character files - Does character background explain this knowledge/action?
4. Common sense - Is there an obvious implied mechanism (e.g., "lawyer knows case details" = standard briefing)?

If Mechanism is Unclear:

• ❌ DO NOT INVENT a mechanism
• ✅ FLAG IT in your context log as "UNVERIFIED BEAT"
• ✅ Ask user for clarification before proceeding
• ✅ Check if mechanism should be shown on-screen vs. implied

Example Verification:

Outline Beat: "Eddie is there, Hank is there"
❓ HOW does Eddie know to come?
✅ VERIFIED: notes.md says "one phone call" → Doogan called Eddie
✅ MECHANISM: Police procedure allows one call, Doogan used it

Outline Beat: "Hank mentions money too many times"
❓ WHY does Hank bring up money?
✅ VERIFIED: Character type = "parking meter with law degree"
✅ MECHANISM: Hank's personality (transactional, business-focused)

Outline Beat: "Eddie knows about Annabelle investigation"
❓ HOW does Eddie know investigation details?
✅ VERIFIED: Episode 1, Scene 8 - Doogan called Eddie from store
✅ MECHANISM: Phone call established information flow

Document Verified Mechanisms: Add to your continuity notes:

• "Eddie knows Doogan was arrested → Doogan called him (one phone call, police procedure)"
• "Hank knows case basics → Eddie briefed him when calling at 3 AM (implied off-screen)"

7. Load ENTIRE Episode Outline & Create Scene Map (MANDATORY - CRITICAL)

BEFORE writing ANY scene, you MUST:

1. Read the COMPLETE episode outline from elements/outlines/episode-##.md
2. Create a SCENE MAP of the entire episode
3. Identify THIS scene's "Button" (ending point) - this is WHERE YOU MUST STOP
4. Identify the NEXT scene's opening - this is what you CANNOT write

Why This Is Non-Negotiable:

• Prevents scene spillover - Most common violation
• Ensures you know what comes BEFORE and AFTER this scene
• Enforces precise scene boundaries (stop at the Button)
• Episode structure dictates pacing - don't break it

CRITICAL RULE: STOP AT THE BUTTON

Every scene in the outline has a "Button:" marker. This is the EXACT point where you MUST stop writing. Do not continue past it. Do not write "what happens next." That's the next scene's job.

Example Violation:

Scene 10 Outline Button: "They have the golden ticket; Jillian is still stuck inside"

❌ WRONG: Write them running back, arriving at country club, seeing Jillian there first
✅ RIGHT: End outside speakeasy with ticket, Jillian still inside, start running back

Scene 12 is "Finish Line" - that's where arrival happens. Don't steal it.

SCENE MAP FORMAT:

EPISODE 2 SCENE MAP:
Scene 1: Holding Cell Graffiti (Doogan POV - interrogation)
  Button: "You've got a lawyer"
  
Scene 2: The Walkout (Doogan POV - lawyer arrives)
  Button: Eddie says "Abby's making pancakes"
  
Scene 3: Eddie's Kitchen (Doogan POV - team briefing)
  Button: Doogan goes home to sleep
  
Scene 4: Doogan's Condo (Doogan POV - shower, sleep, reset)
  Button: Doogan leaves for country club
  
Scene 5: Restaurant Meeting (Doogan POV - client meeting)
  Button: Doogan goes home
  
Scene 6: Country Club Kickoff (Doogan POV - hunt begins)
  Button: "They head for Rose Shop—Jillian tailing"
  
Scene 7: Strip Mall Clinic (Eddie POV - Martinez confrontation)
  Button: Eddie waits for Martinez to leave
  
Scene 8: The Flower Shop (Doogan POV - roses clue)
  Button: "They head for bookstore—Jillian following"
  
Scene 9: Eddie's House (Eddie/Abby/Baxter POV - plan infiltration)
  Button: "Baxter in IT uniform"
  
Scene 10: Bookstore to Speakeasy (Doogan POV - accelerated hunt)
  Button: [to be determined]

CRITICAL VERIFICATION (for Scene 8 example):

WRITING SCENE 8: The Flower Shop

PRIOR SCENE (Scene 7):
- Eddie at clinic
- Martinez confrontation
- Different timeline/POV

THIS SCENE'S BEATS (Scene 8 outline):
- Arrive at Rose Shop
- Jillian follows and eavesdrops
- Cheryl orders red roses (wrong)
- Baxter reads sonnet, identifies "red and white"
- Doogan orders red and white roses
- Clerk gives envelope
- Next clue is Pride and Prejudice
- Baxter finds bookstore
- Cheryl decides to "crush" not just win

THIS SCENE'S BUTTON:
"They head for the bookstore—Jillian and her boyfriend still tailing"

NEXT SCENE (Scene 9):
- Eddie's House
- Different timeline/POV
- Baxter/Abby planning

SCENE AFTER THAT (Scene 10):
- Bookstore to Speakeasy
- THIS is where bookstore happens

BOUNDARY ENFORCEMENT:
❌ DO NOT write: Arriving at bookstore
❌ DO NOT write: Inside bookstore
❌ DO NOT write: Finding Pride and Prejudice clue
❌ DO NOT write: Next clue reveal
✅ STOP AT: Heading toward bookstore

MANDATORY PRE-WRITING CHECKLIST:

Before writing Scene N:

• Read ENTIRE episode outline (all scenes)
• Created scene map with all Buttons identified
• Identified Scene N-1 (what came before)
• Identified Scene N's beats and Button
• Identified Scene N+1 (what comes after)
• Identified Scene N+2 (to catch multi-scene threads)
• Verified NO overlap with Scene N+1 or N+2
• Documented where to STOP writing

If You Cannot Create Complete Scene Map:

• ❌ DO NOT proceed with writing
• ✅ FLAG: "Incomplete episode outline - cannot verify boundaries"
• ✅ Ask user for complete episode structure

8. Compare Outline to Scene (ANTI-FABRICATION CHECK + BOUNDARY ENFORCEMENT)

AFTER creating scene map, do line-by-line comparison:

Read the episode outline beat list for THIS scene:

• List every action mentioned in the outline
• List every piece of dialogue mentioned
• List every object/prop mentioned
• List every character interaction specified
• IDENTIFY WHERE THIS SCENE ENDS (the "Button" in the outline)

CRITICAL: SCENE BOUNDARY RULE

Each scene must END at its designated outline "Button." NEVER spill into the next scene's beats.

Why This Matters:

• Episode outlines define clear scene boundaries
• Each scene has a designated purpose, beats, and ending
• Spillover into next scene creates duplication and ruins pacing
• Next scene's opening is carefully crafted—don't steal it

Scene Boundary Verification:

1. Read THIS scene's outline beats (Scene N)
2. Read NEXT scene's outline beats (Scene N+1)
3. Identify the dividing line:
   • Where does Scene N's "Button" end?
   • Where does Scene N+1's opening begin?
4. STOP writing at Scene N's button—even if it feels like the story should continue

Example (Episode 2):

Scene 6 Outline:
- Terry introduces Doogan to Cheryl
- Rules announced (scavenger hunt, literature theme)
- First clue given: Shakespeare sonnet
- Baxter identifies "roses"
- Button: "They head for Rose Shop—Jillian tailing"

Scene 7 Outline:
- [Different timeline: Eddie at clinic]

Scene 8 Outline:
- Doogan and Cheryl at Rose Shop
- Roses clue solved
- Next clue received

❌ WRONG (Scene 6 spillover):
Scene 6 includes: heading to Rose Shop → arriving → entering → solving clue → leaving
RESULT: Scene 8 duplicates Rose Shop visit

✅ CORRECT (Scene 6 boundary respected):
Scene 6 includes: rules → clue → Baxter identifies location → Button: heading toward Rose Shop
Scene 8 includes: FULL Rose Shop visit (arrival → entry → solve → exit)

Enforcement Checklist:

• Identified where THIS scene's Button ends (per outline)
• Verified NEXT scene's opening (per outline)
• Stopped writing at Button—did not continue into next scene
• If tempted to continue, STOP and flag as "Scene Boundary Risk"

Create "Outline Implementation Map":

Outline Beat: "Terry introduces Doogan to Cheryl"
Implementation: Terry calls Doogan over, names him, Cheryl responds
Verification: ✅ Matches outline exactly

Outline Beat: "First prize is a designer handbag"
Implementation: Terry announces "vintage Birkin handbag"
Verification: ❌ FABRICATION - outline says "designer handbag" (generic), not specific brand
FIX NEEDED: Use generic "designer handbag" OR verify with user which brand

CRITICAL ANTI-FABRICATION RULES:

DO NOT add actions not in outline:

• ❌ Outline: "Baxter feeds location" → ❌ WRONG: Add phone call charade
• ✅ Outline: "Baxter feeds location" → ✅ RIGHT: Baxter speaks in earpiece, Doogan relays info

DO NOT add dialogue beats not in outline:

• ❌ Outline: Silent beat → ❌ WRONG: Add witty banter
• ✅ Outline: Silent beat → ✅ RIGHT: Action/movement only

DO NOT upgrade generic terms to specific brands:

• ❌ Outline: "designer handbag" → ❌ WRONG: "Chanel handbag"
• ✅ Outline: "designer handbag" → ✅ RIGHT: "designer handbag" (or ask user for specific brand)

DO NOT attribute knowledge to characters they shouldn't have:

• ❌ Outline: "Baxter identifies Sonnet 130" → ❌ WRONG: Doogan recognizes it first
• ✅ Outline: "Baxter identifies Sonnet 130" → ✅ RIGHT: Baxter provides info, Doogan relays

DO NOT invent character traits from social class assumptions:

• ❌ ❌ WRONG: "Wealthy character → must know literature" OR "Wealthy character → failed school"
• ✅ ✅ RIGHT: Check character file for actual education/knowledge

IMPLEMENT EXACTLY WHAT'S WRITTEN:

• Outline says character "taps ear" → They tap their ear (not make phone call)
• Outline says "Baxter feeds location" → Baxter speaks in earpiece (already established device)
• Outline says "designer handbag" → Use that exact phrase (or ask which brand)
• Outline says beat is character's knowledge → That character provides it (not POV character)

If outline beat seems incomplete or unclear:

• ❌ DO NOT improvise to "make it work"
• ✅ FLAG IT and ask user for clarification
• ✅ Implement literally what IS written, skip what ISN'T

8. Flag Missing Details (NEW - CRITICAL)

If ANY detail needed for the scene is NOT in element files, STOP writing.

Create a "❓ MISSING DETAILS - USER INPUT NEEDED" section:

Examples of details to flag:

• Character physical descriptions not in files
• Character backgrounds/history not documented
• Relationship histories not established
• Object descriptions (car models, clothing brands, possessions)
• Location specifics (if setting.md doesn't cover it)
• Character motivations not explained in files
• Technical details (procedures, mechanisms, how things work)

Format for asking:

## ❓ MISSING DETAILS - USER INPUT NEEDED

Before proceeding with Scene X, I need clarification:

### 1. [Detail Category]
**Outline says:** [what outline mentions]
**Currently missing:** [what's not in files]
**Question:** [specific question for user]
**Options (if applicable):** [logical possibilities based on context]

### 2. [Next detail]
...

**Please provide answers, and I'll proceed with accurate details.**

DO NOT:

• ❌ Invent details to fill gaps
• ❌ Guess at character backgrounds
• ❌ Make up object descriptions
• ❌ Assume relationship histories
• ❌ Proceed with scene until user provides details

DO:

• ✅ Stop immediately when detail is missing
• ✅ List ALL missing details before asking
• ✅ Provide context from outline/prior scenes
• ✅ Offer logical options when possible
• ✅ Wait for user response
• ✅ Document user's answers in appropriate element file after scene is complete

McDonald Opening Rule (NON-NEGOTIABLE)

See references/mcdonald-rule.md for full explanation.

NEVER open with:

• ❌ Weather descriptions
• ❌ Atmospheric scene-setting
• ❌ Introspection before dialogue
• ❌ Character arriving at location

ALWAYS open with:

• ✅ Dialogue mid-progress
• ✅ Action already in motion
• ✅ Eavesdropper test: reader overhears conversation in progress

Test: Could an eavesdropper understand what's happening without needing narrator exposition first?

Quick Examples

❌ BAD:

Rain drummed against the windows of Eddie's kitchen. Doogan felt exhausted after the interrogation.

✅ GOOD:

"—so Hank costs money, obviously, guy's a lawyer even if he's a bottom-tier one—"

Stephen King Principles

See references/king-principles.md for full details. Key rules:

1. Discovery-Driven - Character action drives plot (not convenience)
2. Sensory Detail - One specific sensory detail per scene beat
3. Emotional Truth - Character behavior must be honest
4. Cut 10-20% - First draft is too long; remove adverbs, strengthen verbs

POV Discipline (CRITICAL)

This project uses third-person close, one POV per scene.

• Reader knows ONLY what POV character knows
• No head-hopping (switching POV mid-scene)
• No omniscient narrator intrusions
• Off-screen events delivered indirectly (briefing, document, report)

Fletch Rule: Reader learns when protagonist learns.

Continuity Violation Prevention (CRITICAL)

Common continuity errors to avoid:

The Secrecy Rule (NON-NEGOTIABLE - STORY-BREAKING IF VIOLATED)

Doogan's business only works if targets never know they're part of a paid operation.

See elements/checklist.md → "The Secrecy Rule" and elements/notes.md → "The Secrecy Rule" for complete guidance.

Critical Points:

• Targets believe Doogan's interest is genuine
• Cover stories explain Doogan's presence (family friend, coincidence, mutual connection)
• Targets may become skeptical about authenticity, but NEVER discover it's a paid service
• NO character references "hiring," "paying," "job," "babysit," or "service" in relation to Doogan

Example violations to NEVER write:

• ❌ "My parents paid you to babysit me"
• ❌ "Is this your job?"
• ❌ "How much are they paying you?"

Acceptable skepticism:

• ✅ "You're too perfect. This feels like a setup."
• ✅ "My parents set me up with someone exactly like you. That's suspicious."
• ✅ "You're really good at this. Too good."

If target discovers the truth, it's a story-breaking crisis event (has NOT happened in Episodes 1-2).

Invented Details

• ❌ Adding store brand names not mentioned (e.g., "Ralphs" when prior scene said "grocery store")
• ❌ Specifying car makes/models not in character files
• ❌ Describing homes/apartments with details not established
• ❌ Inventing clothing brands, styles, or colors
• ❌ Adding physical traits not in character files (eye color, height, build)

Knowledge Violations

• ❌ Character knowing information they haven't learned yet
• ❌ Character forgetting information they learned earlier
• ❌ POV character knowing things that happened off-screen without being told

Self-Contradictory Fabrications (CRITICAL - NEW)

The "cash under the table + tax forms" error class:

These are invented details that directly contradict the mechanism that makes another detail possible.

Examples of self-contradictory fabrications:

• ❌ "Paid cash under the table" + "Baxter found tax records" (tax records mean it WAS reported, not under the table)
• ❌ "Completely off the grid" + "Found through social media" (can't be both)
• ❌ "Never left fingerprints" + "Matched fingerprints to scene" (contradictory)
• ❌ "No paper trail" + "Traced through business filings" (filings ARE paper trail)
• ❌ "Anonymous transaction" + "Credit card receipt" (credit cards aren't anonymous)

Prevention Protocol:

BEFORE writing any "how we found X" explanation:

1. Read the prior scenes that established HOW it was found
2. Identify the mechanism explicitly stated (tax forms, social media, witness statement, etc.)
3. Do not add contradictory color that undermines the mechanism
4. Verify logic chain: If A enabled us to find B, then A must be possible/true

Example from actual error:

❌ WRONG:
"She worked there cash under the table, no official records except the tax forms Baxter pulled"

PROBLEM: "Cash under the table" means NO official reporting
BUT: "tax forms" means it WAS officially reported
THESE CONTRADICT EACH OTHER

✅ CORRECT (honors the mechanism):
"She worked there eight months, filed a 1099 which is how Baxter found the place"

OR (if truly under the table):
"She worked there eight months under the table, but Baxter traced her through [different mechanism]"

BUT NOT BOTH "under the table" AND "tax forms" - pick ONE and stick with it.

Mechanism Verification Checklist:

Before writing ANY "how we know this" or "how we found this" line:

• Read outline: What does it say about how this was discovered?
• Read prior scenes: Was the mechanism already established?
• Check notes.md: Are there procedures or research methods documented?
• Logic check: Does my explanation contradict the mechanism?
• Simplify: Can I just state the mechanism without adding "color" that contradicts it?

If you're unsure about the mechanism:

• ✅ Use the EXACT wording from outline/prior scene
• ✅ Keep it simple: "Baxter found her through tax records" (don't elaborate)
• ✅ Flag for user review if mechanism seems unclear

RED FLAGS that indicate potential self-contradiction:

• Writing "but" or "except" in same sentence as contradictory detail
• Adding "color" details about how something was done without checking if that contradicts the discovery method
• Assuming details about financial arrangements (cash/check/wire/reported/unreported) without checking canon
• Embellishing with specific methods when outline only says "found" or "traced"

Relationship Errors

• ❌ Changing established relationship dynamics without cause
• ❌ Characters using nicknames not yet established
• ❌ Ignoring established tensions or conflicts

Physical Continuity

• ❌ Objects appearing/disappearing (if character arrived with coffee cup, it stays unless mentioned)
• ❌ Time inconsistencies (scene 2 can't be morning if scene 1 ended at night)
• ❌ Location errors (if they're in Eddie's kitchen, stay consistent with established layout)

Rule: When in doubt, stay VAGUE. "She drove away" not "She drove away in her BMW." Generic is safe.

Environment as Character Exposition (CRITICAL)

Use visible objects in a character's environment to convey background, accomplishments, lifestyle, and health status. NEVER have characters talk about their own résumé.

The Rule: POV character observes objects → Infers information → Reader learns through observation

Examples:

❌ WRONG: Career Exposition in Dialogue

"I've been doing this seventeen years."
"Now I manage north of two hundred million."
"Started on the floor of the exchange."

✅ RIGHT: Environment Shows Background

"Ironman triathlon finish shot, sail boat, golf tournament trophy, shots of the man with his trophy family."
— POV character (Eddie) observes photos on credenza
— Reader infers: healthy, active, affluent, accomplished
— Zero dialogue about career/accomplishments

What to Show Through Environment:

• Health: Sports trophies, race medals, athletic photos, gym equipment
• Wealth: Office quality, artwork, furniture, neighborhood, vehicle
• Background: Diplomas, certificates, family photos, memorabilia
• Personality: Desk organization, book selection, décor style, personal items

POV Character's Role:

• Notices specific objects ("Ironman finish photo")
• Makes wry observations ("Outdoor sports?")
• Infers without asking directly
• Lets reader draw conclusions

FORBIDDEN:

• ❌ Characters stating years of experience
• ❌ Characters listing credentials or degrees
• ❌ Characters announcing income/assets managed
• ❌ Characters describing their own accomplishments
• ❌ POV asking "how long have you been doing this?"
• ❌ POV asking "do you exercise much?"

Why This Matters:

• Respects reader intelligence (show, don't tell)
• Keeps dialogue focused on scene's actual purpose
• Creates natural characterization through observation
• Prevents exposition dump interruptions
• Maintains third-person close POV discipline

Scene-Writing Application: When POV character enters a space, describe 2-4 specific visible objects that reveal character background. Let POV character notice them, possibly comment wryly, then move on to scene's actual purpose.

Character Worldview Over Biography (CRITICAL)

One-scene characters must be PEOPLE with worldviews and ethics, not résumés with dialogue. Reveal character through philosophy/values, NEVER through career stats.

The Rule: Character's ETHICS and WORLDVIEW define them → Creates dramatic content → Biography emerges as byproduct

Examples:

❌ WRONG: Career Biography

"Seventeen years in finance."
"Manage portfolios north of two hundred million."
"Clients across three states."

PROBLEM: This is a résumé, not a person. Creates no dramatic tension. Just data delivery.

✅ RIGHT: Character Philosophy/Worldview

"There are two types of investors. Predators and prey. Predators see the shift coming before anyone else. They don't wait for consensus. They move."

REVEALS: Ethics (social-Darwinist), personality (aggressive), worldview (competitive), potential conflict with Eddie's values

What Worldview Reveals:

• Ethics: How they view right/wrong, rules, other people
• Values: What they prioritize (money, status, fairness, loyalty)
• Personality: Aggressive, cautious, cynical, optimistic
• Potential Conflict: Where they might clash with protagonist

How POV Character Responds:

• Reacts to the PHILOSOPHY, not the information
• Makes wry observation ("Outdoor sports?")
• Engages with the worldview or deflects
• Learns information as BYPRODUCT of interaction

Scene Structure:

1. Character states philosophy/worldview
2. POV character responds to WHO this person is
3. Information emerges naturally through interaction
4. Scene has dramatic content (clash of values, personality friction)

FORBIDDEN:

• ❌ Characters stating years in profession
• ❌ Characters listing credentials
• ❌ Characters announcing revenue/clients/accomplishments
• ❌ POV asking "how long have you been doing this?"
• ❌ POV asking about career history
• ❌ Treating one-scene characters as information sources only

Why This Matters:

• Creates actual dramatic content (personality, ethics, potential conflict)
• Respects that even one-scene characters are PEOPLE
• Makes scenes ABOUT something beyond data delivery
• Generates natural dialogue instead of interview format
• Reveals information as byproduct, not purpose

Scene-Writing Application: When introducing any speaking character (even one-scene), establish their worldview/philosophy within first 3-5 lines. Let POV character respond to WHO they are. Information about their role/background emerges through interaction, not interrogation.

Name Every Speaking Character (CRITICAL)

Every character who speaks must have a name. NEVER use generic labels like "the broker," "the nurse," "the clerk." Naming creates presence and respect.

The Rule: If character speaks → Character has name → Use it consistently

Examples:

❌ WRONG: Generic Label

"The broker leaned back."
"The finance guy nodded."
"He gestured to the chart."

PROBLEM: Reduces person to function. Creates distance. Feels like NPC in video game.

✅ RIGHT: Named Character

"Phil Blakely leaned back."
"Blakely nodded."
"He gestured to the chart."

REVEALS: This is a PERSON who exists in this world. Has identity beyond role.

Naming Conventions:

• First Introduction: Full name ("Phil Blakely")
• Subsequent References: Last name ("Blakely") or pronoun ("he")
• Dialogue Tags: Use name in first tag, then pronouns if clear

How to Generate Names:

1. Check if character file exists in elements/characters/
2. If not, check notes.md for any mentioned names
3. If creating new one-scene character, choose name that:
   • Fits setting/genre (professional environment → professional name)
   • Isn't similar to existing character names
   • Feels real (not placeholder like "Bob Smith")
   • Has appropriate ethnicity/cultural background for setting

FORBIDDEN:

• ❌ "the broker," "the finance guy," "the man"
• ❌ "the nurse," "the receptionist," "the clerk"
• ❌ "the waiter," "the driver," "the trainer"
• ❌ Using profession as label for speaking characters
• ❌ Leaving character unnamed because they're "minor"

Exception: Non-speaking background characters CAN remain unnamed ("a waiter refilled water glasses"). But if they speak, they need a name.

Why This Matters:

• Shows respect for every character as a person
• Creates sense of populated, real world
• Prevents "NPC" feeling in scenes
• Makes dialogue tags clearer and more natural
• Establishes that even one-scene characters matter

Scene-Writing Application: Before writing ANY scene with new speaking character, determine their name. Add to notes.md after scene complete if they might appear again. For true one-scene characters, name is sufficient—no need to create full character file unless they become recurring.

Every Scene Must Be ABOUT Something (CRITICAL)

Scenes are dramatic encounters between people, NOT information-delivery mechanisms. The scene must be ABOUT character interaction, worldview clash, or emotional stakes—information emerges as byproduct.

The Rule: Scene's PURPOSE ≠ Delivering information Scene's PURPOSE = Dramatic content (clash, tension, personality, decision) Information = Byproduct of dramatic interaction

Examples:

❌ WRONG: Scene as Data Delivery

Purpose: "Eddie needs to learn if broker is healthy"
Structure: Eddie asks health questions → Broker answers → Eddie gets information → Scene ends
Result: Interview format. No dramatic content. Feels mechanical.

✅ RIGHT: Scene as Dramatic Encounter

Purpose: "Eddie encounters predatory finance-bro philosophy and responds to WHO this person is"
Structure: Blakely states worldview (predators/prey) → Eddie responds with wry observation → Eddie infers health from environment → Scene is ABOUT the personality clash
Byproduct: Eddie learns broker is healthy (through photos, trophies)
Result: Actual dramatic content. Character-driven. Information emerges naturally.

What Makes Scene "ABOUT Something":

• Personality Clash: POV character encounters different worldview/ethics
• Values Conflict: Characters have competing priorities
• Emotional Stakes: Decision, revelation, or realization with weight
• Character Choice: Someone makes decision that reveals who they are
• Power Dynamic: Status, control, or influence shifts between characters

Scene Structure (Proper):

1. Opening: Drop into existing interaction (McDonald rule)
2. Dramatic Content: Character philosophies, ethics, personalities interact
3. POV Response: Protagonist reacts to WHO the other person is
4. Information Byproduct: Necessary plot info emerges through interaction
5. Button: Scene ends on turn or decision, propels to next scene

FORBIDDEN:

• ❌ Scene exists ONLY to deliver plot information
• ❌ POV character asks direct questions to extract data
• ❌ Other character answers in exposition mode
• ❌ Dialogue feels like interrogation or interview
• ❌ No actual dramatic content (personality, conflict, decision)
• ❌ Scene could be replaced with "Eddie researched the broker and learned he was healthy"

The "Could This Be a Paragraph?" Test: If scene could be replaced with single paragraph of summary without losing anything important, scene ISN'T about something—it's just data delivery. Fix it.

Why This Matters:

• Respects reader expectation (scenes are for drama, not just info)
• Creates actual story content (characters, conflict, stakes)
• Makes protagonist ACTIVE (responding to people, not extracting data)
• Generates organic dialogue (interaction, not interrogation)
• Information lands better when it's byproduct of drama

Scene-Writing Application: BEFORE writing scene, ask: "What is this scene ABOUT (dramatically)?" If answer is only "character learns X information," stop and redesign. Find the dramatic content: whose worldview appears? What personality does POV encounter? What's the actual human interaction? Then let inf

…(truncated)