VERITY — AI Behavioral Analysis Tool

A forensic linguistic pattern analyzer that detects deception markers in text using Claude AI.

---

What It Does

VERITY analyzes text for evidence-based behavioral deception patterns — it doesn't claim to "detect lies" magically. Every flag comes with an explanation of why it's suspicious, grounded in real linguistic research (Statement Validity Analysis, SCAN methodology, cognitive load research).

Three Modes

Mode	Description
ANALYZE	Paste/type/speak text. Get a deception score, annotated highlights, and a full breakdown.
TRAINING	6 real-world scenarios. Guess truthful/deceptive before the reveal. Scored at the end.
INTERROGATION	Play detective. Question an AI suspect (guilty or innocent). Make your accusation.

---

Tech Stack

Layer	Technology
Framework	Next.js 14 (App Router)
Language	TypeScript
Styling	Tailwind CSS + custom CSS
Animations	Framer Motion
Icons	Lucide React
AI	Anthropic Claude API (claude-opus-4-6 for analysis, claude-sonnet-4-6 for interrogation)
Speech	Web Speech API (browser native)

---

Getting Started

1. Clone and install

git clone <your-repo>
cd ai-lie-detector
npm install

2. Set up environment

cp .env.example .env.local

Edit .env.local and add your Anthropic API key:

ANTHROPIC_API_KEY=sk-ant-your-key-here

Get an API key at console.anthropic.com.

3. Run development server

npm run dev

Open http://localhost:3000.

---

Project Structure

ai-lie-detector/
├── app/
│   ├── layout.tsx              # Root layout, metadata, font import
│   ├── page.tsx                # Main page with mode routing
│   ├── globals.css             # CSS variables, animations, base styles
│   └── api/
│       ├── analyze/route.ts    # POST /api/analyze — forensic text analysis
│       └── interrogate/route.ts # POST /api/interrogate — suspect roleplay + analysis
│
├── components/
│   ├── Header.tsx              # Top nav with mode tabs
│   ├── AnalyzeMode.tsx         # Analyze mode orchestrator
│   ├── InputPanel.tsx          # Text input + speech recognition
│   ├── ResultsPanel.tsx        # Analysis results layout
│   ├── DeceptionGauge.tsx      # Animated SVG arc gauge
│   ├── PatternHighlighter.tsx  # Text with color-coded highlights + tooltips
│   ├── BreakdownPanel.tsx      # Flagged patterns + linguistic profile
│   ├── TrainingMode.tsx        # Scenario-based training module
│   ├── InterrogationMode.tsx   # Detective/suspect chat game
│   ├── LoadingState.tsx        # EKG animation during analysis
│   └── EmptyState.tsx          # Pre-analysis landing content
│
├── hooks/
│   └── useSpeech.ts            # Web Speech API hook
│
└── lib/
    ├── types.ts                # All TypeScript interfaces
    ├── constants.ts            # Training/interrogation scenarios, pattern metadata
    └── utils.ts                # Text segmentation, score helpers, formatting

---

🔬 Detected Deception Patterns

Pattern	Description
Hedging	"maybe", "I think", "sort of", "I believe"
Qualifier	"honestly", "to be honest", "I swear", "believe me"
Distancing	"that" vs "this", passive voice, third-person self-reference
Over-Explanation	Unnecessary detail, unprompted justifications
Formal Denial	"I did not" vs "I didn't" — unnatural in casual speech
Tense Shift	Switching to present tense mid-story (common in fabrication)
Inconsistency	Contradictory statements within the same passage
Emotional Flatness	Absence of expected emotional language
Filler Words	"um", "uh", "like", "you know" (cognitive load markers)
Memory Qualification	Selective "I can't remember" on key moments

---

API Reference

POST /api/analyze

// Request
{ "text": "string (min 20 chars)" }

// Response
{
  "deceptionScore": 73,
  "confidence": 85,
  "verdict": "suspicious",
  "summary": "...",
  "patterns": [
    {
      "id": "abc123",
      "type": "qualifier",
      "phrase": "to be honest",
      "explanation": "...",
      "severity": "high",
      "confidence": 92
    }
  ],
  "keyInsights": ["...", "...", "..."],
  "linguisticProfile": {
    "hedgingDensity": 45,
    "qualifierUsage": 80,
    "negationPatterns": 30,
    "temporalConsistency": 60,
    "emotionalAuthenticity": 40
  }
}

POST /api/interrogate

// Request
{
  "scenario": { InterrogationScenario },
  "messages": [ InterrogationMessage[] ],
  "newMessage": "string"
}

// Response
{
  "response": "Suspect's reply...",
  "analysis": {
    "deceptionScore": 68,
    "flags": ["hedging language", "over-qualified"]
  }
}

---

Design System

Palette:

• #060A10 — Background base
• #00CFFF — Cyan (primary accent / truthful-leaning)
• #00E87A — Green (truthful verdict)
• #FF2855 — Red (deceptive verdict)
• #FFB020 — Amber (uncertain)

Fonts:

• Oxanium — Display / scores / verdicts
• DM Sans — Body text
• JetBrains Mono — Data / labels / mono

---

Important Disclaimer

VERITY is a linguistic pattern analysis tool — not a verified lie detection system. Deception analysis based on language patterns alone has significant limitations:

• Nervous people may exhibit deception markers while telling the truth
• Cultural and neurodivergent communication differences affect output
• Short texts yield less reliable results
• This is not admissible as evidence in any legal context

Use for educational purposes, creative writing, research, and entertainment only.

---

Extending the App

Add more training scenarios → lib/constants.ts — TRAINING_SCENARIOS array

Add more interrogation cases → lib/constants.ts — INTERROGATION_SCENARIOS array

Tune the analysis prompt → app/api/analyze/route.ts — SYSTEM_PROMPT

Change the models → Both API routes have model strings at the top

---

License

MIT — use freely, modify boldly.