90day-voice-gate

name

90day-voice-gate

description

Quality filter that catches AI-sounding language and enforces TLC voice standards before content reaches EPs or network. Use this skill whenever reviewing beat sheets, scene expansions, or any production writing for voice compliance. Triggers on "voice check", "quality gate", "EP ready", "sounds too AI", "too verbose", "slop check", "voice pass", "clean this up", "network ready", "before we send", or any request to review/polish production content before submission. Also use proactively as a final pass after 90day-story-development or 90day-scene-refinement completes work.

90 Day Voice Gate

The last filter before production writing reaches executive producers. This skill exists because AI-generated content carries a specific scent — verbosity, hedging, over-explanation, symmetrical sentence structures — that trained EPs recognize immediately. The goal is to strip that scent completely so the output reads like it was written by a story producer who's been on the franchise for five seasons.

Think of this as an immune system. It recognizes foreign patterns (AI tells, voice violations, formatting drift) and eliminates them before they enter the bloodstream (EP review).

When to Use This Skill

• Before EP submission — Run every piece of production writing through this gate
• After story-development output — As the final polish step in the pipeline
• After scene-refinement — Verify the trimmed version still holds voice
• When feedback says "sounds off" — Diagnose exactly what's wrong and fix it
• Proactive quality pass — Before any content leaves your hands

The Voice Gate Process

PASS 1: AI Pattern Detection (Strip the Slop)

Scan the entire document for these AI writing fingerprints. Read references/ai-slop-patterns.md for the comprehensive pattern list. The most common offenders:

Verbal Hedging

• "seems to", "appears to", "begins to" — Cut these. Characters DO things.
• "a sense of", "a feeling of", "a moment of" — Nominalization. Kill it.
• "clearly", "obviously", "notably" — Filler adverbs that add nothing.

Over-Explanation

• Sentences that explain what the previous sentence already showed
• Emotional state described AND demonstrated (pick one — usually the demonstration)
• Setup → Action → Reaction → Reflection chains where Action + one other suffices

Symmetrical Structure

• AI loves balanced pairs: "On one hand... on the other hand"
• Parallel constructions repeated 3+ times in sequence
• Lists of three where two would hit harder
• Every paragraph opening with the same structure

Connector Bloat

• "However", "Furthermore", "Additionally", "Moreover" — These are essay connectors, not TLC voice
• "In this moment" — AI favorite. Cut every time.
• "As they continue to" — Process narration. Cut.
• "It becomes clear that" — Telling, not showing. Cut.

Passive Voice Infiltration

• "is seen", "can be felt", "is revealed" — Active voice always: "We see", "She reveals"
• "There is a tension" — No. "Tension builds" or better yet, show the tension through action.

PASS 2: TLC Voice Enforcement

Check every line against the TLC Voice Signature DNA:

Sentence Architecture Check

• Target: 10-15 words average with intentional variety
• Count a sample of 10 sentences. If average exceeds 18 words, flag for trimming.
• Fragments used for punch? ("Yep — he has a job!")
• Questions driving narrative forward? ("Will she agree?")

Action Language Check Compare against approved patterns:

• APPROVED: "Jenny shows off their new apartment and comes across Sumit unpacking"
• APPROVED: "Manuel enters in a construction uniform and gives her a little cat call"
• NOT APPROVED: "Jenny is happy as she shows the apartment" (emotional telling)
• NOT APPROVED: "Manuel walks in wearing his uniform" (flat, no energy)

The difference: approved patterns have ENERGY. Characters show off, come across, enter and give. Not "is happy", not "walks in wearing."

Dialogue Treatment Check

• Zero direct quotes? (Non-negotiable)
• Topics described with emotional intent? ("confesses", "demands", "admits" — not "says", "tells", "mentions")
• Questions used to drive uncertainty? ("How does he respond?")

Tense Check

• Present tense throughout verite? (Scan for -ed endings, "was", "had been")
• Past tense ONLY in INTV backstory sections?

Comedy Check

• At least one comedy beat suggested per scene?
• Phrased as questions? ("Does she kiss her bicep?")
• Not forced into crisis moments?

PASS 3: Formatting Compliance

This is about matching what network actually sees — not what looks good in a draft.

Structure Check

• Verite runs 2-4 paragraphs before INTV interrupts?
• INTV clearly marked as separate block with "In INTV" or "In INTERVIEW" marker?
• Scene word count in 600-800 range?
• Balance roughly 80% verite / 15% OTF / 5% INTV?

Movement Mandate Check

• Every conversation wrapped in activity?
• No static scenes? (Scan for: sitting, standing, looking at each other without doing anything)
• Physical comedy opportunities identified?

Self-Shooting Check

• Camera position established at scene open?
• Every shot achievable by cast with phone/tripod?
• No impossible angles or crew-required setups?

PASS 4: Before/After Transformation

For every flagged issue, provide:

1. The exact line or passage that fails
2. Which rule it violates (with rule number from Golden Rules)
3. The transformed version
4. One line explaining why the fix works

Output Format

Structure your voice gate report like this:

VOICE GATE REPORT — [Couple Name] [Scene/Document Title]

OVERALL STATUS: [PASS | PASS WITH NOTES | REVISION REQUIRED | MAJOR ISSUES]

---

AI PATTERN FLAGS: [count]
[List each with line, violation, fix]

VOICE VIOLATIONS: [count]
[List each with line, rule violated, fix]

FORMAT ISSUES: [count]
[List each with specific problem and fix]

---

SENTENCE METRICS:
- Average sentence length: [X] words (target: 10-15)
- Longest sentence: [X] words — [quote it]
- Present tense compliance: [X]%
- Direct quote violations: [count]

---

REVISED VERSION:
[Full cleaned version with all fixes applied]

---

CHANGES MADE: [count]
[Numbered list of every change, grouped by type]

Scoring Thresholds

• PASS — 0-2 minor flags, all easily fixed. Voice reads clean.
• PASS WITH NOTES — 3-5 flags, mostly AI pattern cleanup. Structure is solid.
• REVISION REQUIRED — 6+ flags or any Golden Rule violation. Needs another pass.
• MAJOR ISSUES — Structural voice problems. Multiple Golden Rule violations. Likely needs scene-refinement first.

Integration with Other Skills

• After 90day-story-development: Run voice-gate on Phase 4 output before it goes anywhere
• After 90day-scene-refinement: Verify trimmed version maintained voice quality
• Before 90day-production-readiness: Clean voice ensures accurate rubric scoring
• Feeds back to 90day-scene-refinement: If voice-gate finds structural bloat, route back for refinement first

The North Star

Read the approved samples in the knowledge base (Jenny/Sumit, Ashley/Manuel, Colt/Debbie voice patterns). These are the frequency your output needs to match. When in doubt, read an approved passage aloud, then read your output aloud. If they don't sound like they came from the same writer, keep editing.

The EPs know what network-ready sounds like. They've read thousands of beat sheets. One "furthermore" or one "it becomes clear that" and the whole document gets flagged. The voice-gate catches those before they do.