ARGOS — Manipulation Detection Engine

"Is this manipulation?" — answered in seconds, with philosophical precision.

ARGOS analyzes text through 11 philosophical frameworks and a Gemini AI semantic layer to detect persuasion and manipulation techniques. No prompt engineering. No black box. Paste text, get a verdict.

---

Why ARGOS instead of just asking an AI?

When you ask ChatGPT "is this manipulative?", you get a paragraph. ARGOS gives you:

• A structured verdict: MANIPULATIVE / PERSUASIVE / MIXED / CLEAN
• A score with confidence level
• A breakdown by philosophical framework — not just what is manipulative, but why, through the lens of Aristotle, Foucault, Kant, Orwell, and 7 others
• Reproducible results — the 11 rule-based detectors always give the same answer for the same text
• Evidence: exact phrases flagged as signals, per detector

---

The 11 Philosophical Frameworks

#	Philosopher	Framework	What it detects
1	Aristotle	Rhetorical Appeals	Imbalance between ethos, pathos, and logos
2	Foucault	Power Dynamics	Asymmetric power language, threats of exclusion
3	Bakhtin	Polyphony & Voice	Monologism, absolutist language, propaganda patterns
4	Goffman	Frame Analysis	Emergency framing, false consensus, catastrophism
5	Searle	Speech Acts	Directives disguised as facts, unverifiable promises
6	Toulmin	Argument Structure	Claims without data, manipulation through bare assertion
7	Lakoff	Conceptual Metaphors	War/disease/cleansing metaphors that constrain thought
8	Carnegie	Interpersonal Manipulation	False familiarity, flattery, artificial urgency
9	Kant	Categorical Imperative	Instrumentalizing people, bypassing rational autonomy
10	Habermas	Communicative Action	Strategic action disguised as dialogue, consensus faking
11	Orwell	Politics & Language	Euphemisms, passive voice erasing agency, weasel words

Plus a Gemini 2.5 Flash semantic layer that catches what exact-match rules miss: paraphrase, morphological variants, implicit threats, and informal constructions.

---

Real Use Cases

• Journalists — analyze political speeches and press releases
• Researchers — study propaganda and rhetorical techniques at scale
• Educators — teach critical thinking with concrete, labeled examples
• Individuals — understand why an email or message feels "off"
• Developers — integrate via API for content moderation pipelines

---

Tech Stack

• Backend: FastAPI (Python) — deterministic fractions.Fraction arithmetic for rule detectors
• Frontend: React + TypeScript + Tailwind + Framer Motion
• ML layer: Google Gemini 2.5 Flash via AI Studio API
• Deployment: Vercel (serverless Python + static build)
• Languages: English and Spanish (bilingual detection and UI)

---

Architecture

Text input
    │
    ├── 11 Rule-Based Detectors (deterministic, no API calls)
    │       Aristotle · Foucault · Bakhtin · Goffman · Searle
    │       Toulmin · Lakoff · Carnegie · Kant · Habermas · Orwell
    │
    └── Gemini 2.5 Flash (semantic layer — catches what rules miss)
                │
                ▼
        Weighted Aggregate Score
        (only fired detectors count — silent ones don't dilute signal)
                │
                ▼
        Verdict + Score + Confidence + Per-detector breakdown

Scoring logic: only detectors that fire (score ≥ 1/10) contribute to the weighted average. A convergence bonus is added when ≥ 3 detectors independently flag the same text. Lakoff is contextual only and does not affect the final score.

---

Running Locally

# Backend
pip install -r requirements.txt
GEMINI_API_KEY=your_key uvicorn main:app --reload

# Frontend
cd frontend
npm install
npm run dev

Environment Variables

Variable	Required	Description
GEMINI_API_KEY	Yes	Google AI Studio API key — get one here

---

API

POST /analyze

{
  "text": "You MUST act now — there is no other option."
}

Response:

{
  "verdict": "MANIPULATIVE",
  "score": "9/10",
  "score_float": 0.9,
  "confidence": "HIGH",
  "summary": "Multiple high-confidence manipulation techniques detected...",
  "detectors": [
    {
      "name": "searle",
      "philosopher": "Searle — Speech Acts",
      "score": "11/12",
      "signals": ["HIGH_DIRECTIVE_DENSITY", "FINANCIAL_BAIT_PATTERN"],
      "evidence": ["MUST act now", "no other option"],
      "label": "Commanding and pressuring — text acts as authority instrument"
    }
  ]
}

GET /example

Returns a pre-loaded example text with full analysis.

GET /health

{ "status": "ok", "version": "0.1.0" }

---

Built with

Philosophical rigor. Python fractions. A bit of Gemini.