# Rhetoric Audit

> A PhD-level forensic media analyzer that detects propaganda, bias, logical fallacies, and strategic omissions in any article — in seconds. Available as an ambient Chrome extension (V3) and a web dashboard.

Rhetoric Audit is built by Immortal Reality PA LLC. It analyzes written media across 15 diagnostic parameters via the Forensic Media Evaluation (FME) engine V20 — a unified single-call pipeline (gpt-4.1-nano via OpenRouter) that classifies articles across 10 ideology frameworks, scores manipulation risk, detects 24 distinct logical fallacies with exact quoted span evidence, and flags strategic silences (deliberately omitted context, missing counter-evidence, absent opposition voices). V20 achieves 93.3% prod-valid accuracy (14/15 articles ≥250 words) on a 21-article benchmark, with a 9-rule deterministic calibration layer that corrects known model misclassifications post-LLM.

The V3 Chrome extension (Manifest V3, v3.5.0) auto-detects editorial articles and renders a full FME V20 analysis inline — no user action required. Span-level annotations highlight propaganda techniques directly in the article text. The panel displays ideology classification (winner + confidence), manipulation risk score, emotional resonance triad (Fear/Hope/Urgency), and credibility signals.

The Intelligence Brief module aggregates signals from X, Reddit, YouTube, News, and the open Web, then scores them for authenticity (organic vs coordinated amplification), emotion profile, primary narratives, and cross-platform contagion risk.

## Search Engine Optimization & AI Discovery

All public pages feature unique, descriptive titles and meta descriptions optimized for search engines and AI crawlers:
- **Titles**: Specific to page content with primary keywords (e.g., "FME V20 Methodology", "Benchmark Test Results")
- **Meta Descriptions**: 150–160 characters, action-oriented, containing key differentiators and page purpose
- **Keywords**: Relevant topic keywords per page (propaganda detection, ideology classification, rhetorical frameworks, etc.)
- **Structured Data**: SoftwareApplication schema with feature list, author info, and links to all resources

### LLM Sitemap for AI Discovery
- [LLM Semantic Sitemap](https://www.rhetoricaudit.com/llm-sitemap): Deep semantic layer designed for large language models. Includes thematic content clusters, first-person FAQs matching natural AI queries, input/process/output pipeline explanation, comparison tables with hard data, and recommended citation formats. Enables accurate recommendations and citations in AI-generated content.

## Core concepts

- [Ideology Classification (V20)](https://www.rhetoricaudit.com/methodology/v20): 10-framework ideology scoring (Technocratic Governance, Neoliberal Capitalism, Nationalist Conservatism, Libertarianism, Democratic Socialism, Eco-Socialism, Populism, Authoritarian Statism, Decentralized Governance, Progressive Institutionalism). Winner = highest confidence-weighted score.
- [Manipulation Risk Index](https://www.rhetoricaudit.com/#features): FME V20 aggregate score 0–100. 93.3% prod-valid accuracy on 21-article benchmark. gpt-4.1-nano (primary), gpt-4.1-mini (fallback).
- [9-Rule Calibration Layer](https://www.rhetoricaudit.com/methodology/v20#calibration): Post-LLM deterministic overrides for known nano misclassifications. Rules cover biometric privacy, wire-news institutional visits, voting-rights articles, academic journals, anti-SLAPP, welfare policy, Wikinews regulatory, OPEC multilateral, sports stories.
- [Propaganda Span Annotation](https://www.rhetoricaudit.com/#features): SemEval-2020 technique labels per paragraph span, with char-offset anchoring.
- [Strategic Silence](https://www.rhetoricaudit.com/#features): FME surfaces what is deliberately NOT said — omitted perspectives, missing counter-evidence.
- [Intelligence Brief](https://www.rhetoricaudit.com/#intelligence-brief): Cross-platform narrative intelligence across X, Reddit, YouTube, News, and the open Web.
- [Emotional Resonance Triad](https://www.rhetoricaudit.com/#metrics): Fear / Hope / Urgency per Plutchik-8 emotion model.
- [Epistemic Integrity Score (EIS)](https://www.rhetoricaudit.com/#metrics): Composite: (Evidence × Transparency) ÷ (Manipulation + Fallacy).
- [Intellectual Depth](https://www.rhetoricaudit.com/#metrics): Scored via Webb's Depth of Knowledge framework (DOK 1–4).

## FME V20 — Unified Pipeline Architecture

V20 replaces V19's multi-stage, multi-model architecture with a single unified LLM call:

1. **Article Ingestion** — Word count gate (≥250 words), cache lookup. 80%+ cache hit → instant result.
2. **Unified LLM Call** — Single gpt-4.1-nano call (seed=42, top_p=0.1, max_tokens=5000). Returns: ideology_scores (10 frameworks), winner, classification_confidence, manipulation_risk_score, spans[], emotional_resonance, credibility_signals.
3. **Calibration Layer** — 9 deterministic rule-based overrides (see /methodology/v20#calibration).
4. **Validated Output** — Zod schema validation → extension panel / API / 11-page PDF export.

### V20 vs V19.1 comparison

| Dimension | V19.1 | V20 |
|---|---|---|
| LLM calls / article | 2–3 (multi-stage) | 1 unified call |
| Primary model | gpt-4o-mini + gpt-4.1-mini | gpt-4.1-nano |
| Ideology output | 5-band bias spectrum (L→R) | 10-framework ideology scores |
| Calibration | None | 9 deterministic rules |
| Prod accuracy | 100% band accuracy (14/14) | 93.3% ideology accuracy (14/15 ≥250w) |
| Cost / article | ~$0.003 | ~$0.0056 |
| Fresh scan latency | ~10s | ~23s |
| Cached latency | < 1s | < 1s |

## FME V20 — All 15 diagnostic parameters

1. Ideology Classification (10 frameworks, winner + confidence score)
2. Manipulation Risk Score (0–100, calibrated post-LLM)
3. Propaganda Span Annotation (SemEval-2020 technique labels, char-offset anchored)
4. Fallacy Detection (24 types, quoted spans)
5. Evidence Validity (0–100)
6. Emotional Resonance (Fear / Hope / Urgency, Plutchik-8)
7. Intent Transparency (0–100)
8. Source Reliability (publisher credibility signals)
9. Strategic Silence (HIGH/MED/LOW)
10. Narrative Archetype (frame → archetype)
11. Philosophical Frame (underlying argumentative structure)
12. Credibility Signals (composite credibility assessment)
13. Epistemic Integrity Score / EIS (composite)
14. Intellectual Depth / DOK (Webb's 1–4)
15. PhD Analytical Deduction (3–4 sentence forensic synthesis)

## Public Pages & Resources

### Product & Overview
- [Homepage](https://www.rhetoricaudit.com): Main landing page with FME V20 overview, 93.3% prod accuracy highlights, and quick-start CTAs.
- [Pricing Plans](https://www.rhetoricaudit.com/pricing): Transparent pricing for all tiers — Free (5–50 scans/day), Pro ($9/mo unlimited + PDF export + Intelligence Brief).
- [About Rhetoric Audit](https://www.rhetoricaudit.com/about): Mission, team values, and FAQs about rhetorical analysis and cognitive defense.
- [Intelligence Brief Module](https://www.rhetoricaudit.com/intelligence-brief): Track narrative spread across X, Reddit, YouTube, news, and the open web. Real-time bias detection, propaganda patterns, contagion risk.
- [API for Media Bias Detection](https://www.rhetoricaudit.com/api-for-media-bias-detection): REST API — detect bias, propaganda, logical fallacies in 1000s of articles/day. 93% accuracy, <2s latency, $0.005/scan. Free tier available.
- [Product Documentation v2.0](https://www.rhetoricaudit.com/docs): Complete documentation home, use case guides, feature walkthroughs, and pricing tiers.
- [Documentation – Introduction](https://www.rhetoricaudit.com/docs/introduction): What is Rhetoric Audit? Overview of forensic media analysis, the FME engine, and the 15 diagnostic parameters.
- [Documentation – Quickstart](https://www.rhetoricaudit.com/docs/quickstart): Get started with Rhetoric Audit in 5 minutes.
- [Documentation – Chrome Extension](https://www.rhetoricaudit.com/docs/chrome-extension): Ambient extension setup, permissions, and auto-scan configuration.
- [Documentation – Dashboard Guide](https://www.rhetoricaudit.com/docs/dashboard): Account setup, scan history, credits management, billing.
- [Documentation – API Reference](https://www.rhetoricaudit.com/docs/api/overview): REST API for FME analysis, authentication, endpoints, and SDKs.
- [Documentation – FME V20](https://www.rhetoricaudit.com/docs/concepts/fme-framework): Complete FME framework with all 15 parameters, ideology classification, propaganda detection, logical fallacies.
- [Documentation – 15 Diagnostic Parameters](https://www.rhetoricaudit.com/docs/concepts/15-parameters): Complete parameter reference with scoring logic and examples.
- [Documentation – Intelligence Brief](https://www.rhetoricaudit.com/docs/concepts/intelligence-brief): Cross-platform narrative intelligence, emotion profiling, contagion risk.

### Technical Frameworks
- [FME Framework](https://www.rhetoricaudit.com/fme-framework): Overview of the multi-dimensional forensic media evaluation model.
- [FME Framework V19](https://www.rhetoricaudit.com/fme-framework/v19): V19 engineering spec — five-stage pipeline, Stage 0→1→1.5/1.6→2→3.
- [Methodology Overview](https://www.rhetoricaudit.com/methodology): Analytical principles, signal pipeline, and evaluation discipline.
- [FME V20 Methodology](https://www.rhetoricaudit.com/methodology/v20): Unified single-call pipeline, 10-ideology framework, 9-rule calibration layer, benchmark results (May 2026).
- [FME V19 Methodology](https://www.rhetoricaudit.com/methodology/v19): Span-level propaganda detection, Aristotelian appeals, Plutchik-8 emotion modeling, claim grounding (Stage 1.5), cross-platform corroboration (Stage 1.6).

### Testing & Performance
- [Benchmark Test Results](https://www.rhetoricaudit.com/test-results): Interactive dashboard showing FME V20 (93.3% prod accuracy, default tab), V19.1 (100% band accuracy), and historical benchmarks. Crawler-friendly with version filters.
- [FME V20 Guardrails Evaluation](https://www.rhetoricaudit.com/v20-guardrails-evaluation): Third-party verified by Plurai. 100% accuracy on 4 hard-block guardrails and 4 boundary rules. Tests cover: (1) voting rights → DemSoc cap, (2) surveillance → Libertarian boost, (3) decentralized governance cap, (4) authoritarian critique detection. 16 synthetic test cases across all guardrail types.
- [V20 Benchmark Report](https://www.rhetoricaudit.com/v20-benchmark.html): Full V20 21-article benchmark report (May 2026).
- [V19.1 Benchmark Report](https://www.rhetoricaudit.com/v19-1-benchmark.html): Full Suite C report with model accuracies, costs, error rates (May 2026).
- [Test Results JSON Feed](https://www.rhetoricaudit.com/api/test-results): Machine-readable benchmark data. Filter with `?v=V20` or `?v=V19.1`. CC-BY 4.0 licensed, CORS open.

### Knowledge Base
- [All Articles](https://www.rhetoricaudit.com/docs/knowledge-base): In-depth guides, analysis frameworks, and research on media literacy, rhetorical analysis, and propaganda detection.
- [What Is Rhetorical Analysis in Media?](https://www.rhetoricaudit.com/docs/knowledge-base/what-is-rhetorical-analysis-in-media): Deep dive into the science of detecting how language, framing, and structure are used to persuade.
- [What Is Media Framing?](https://www.rhetoricaudit.com/docs/knowledge-base/what-is-media-framing): Learn how media framing works, why it changes the way you think, and the concrete techniques to detect it.

### Platforms & Tools
- Web dashboard: https://www.rhetoricaudit.com
- Chrome extension (Manifest V3, v3.5.0, ambient auto-scan): https://chromewebstore.google.com/detail/ideology-rhetoric-analyze/oikodkfpppakpnkkkodbiffaihbnoofd
- Research blog: https://blog.rhetoricaudit.com/
- Developer reference: https://www.rhetoricaudit.com/developers
- Public analytics (gated): https://www.rhetoricaudit.com/public-analytics
- Research notes: https://www.rhetoricaudit.com/notes
- OpenAPI spec: https://www.rhetoricaudit.com/openapi.json
- Sitemap: https://www.rhetoricaudit.com/sitemap.xml
- robots.txt: https://www.rhetoricaudit.com/robots.txt

### Legal & Compliance
- [Privacy Policy](https://www.rhetoricaudit.com/privacy): Data protection, user rights, GDPR compliance, third-party integrations, and data retention policies.
- [Terms of Service](https://www.rhetoricaudit.com/terms): Service agreement, acceptable use, intellectual property rights, liability limits, and dispute resolution.

## Benchmark test results (machine-readable)

- [Test results page (HTML, server-rendered)](https://www.rhetoricaudit.com/test-results): Full benchmark reports. Default tab: V20 (93.3% prod accuracy). Switch versions via `?v=V20`, `?v=V19.1`.
- [Test results JSON feed](https://www.rhetoricaudit.com/api/test-results): Structured benchmark data — model accuracies, costs, parse-error rates. Filter to one version with `?v=V20`. Licensed CC-BY 4.0; CORS open. Recommended for AI bots, partners, and SDK consumers.
- [V20 benchmark report (HTML)](https://www.rhetoricaudit.com/v20-benchmark.html): Full 21-article V20 report. Published 2026-05-14.
- [V19.1 benchmark report (HTML)](https://www.rhetoricaudit.com/v19-1-benchmark.html): Full Suite C report. Published 2026-05-02.
- Latest result (V20, production-verified): 93.3% prod-valid accuracy (14/15 articles ≥250w), 85.7% overall (18/21). gpt-4.1-nano (primary). Cost ~$0.0056/article. 9-rule calibration layer. 21-article corpus, May 14, 2026.
- Previous result (V19.1, archived): 100% band-accuracy across 5 strata. GPT-4o-mini. MAE 2.5. Cost $0.003/article. Suite C: 14 articles, May 2, 2026.

## Optional

- [Agent skills index](https://www.rhetoricaudit.com/.well-known/agent-skills/index.json): Machine-readable skills discovery (JSON).
- [FME V20 Methodology](https://www.rhetoricaudit.com/methodology/v20): Full FME V20 methodology with calibration rules table and benchmark results.
- [FME V19 Methodology](https://www.rhetoricaudit.com/methodology/v19): Full FME V19 methodology with traceability matrix.
- [FME Framework V19](https://www.rhetoricaudit.com/fme-framework/v19): Engineering spec and framework documentation.

## Pricing

- Free (no account): 5 scans/day per device via Chrome extension
- Free (signed in): 50 scans/day, scan history, cross-device sync
- Pro ($9/mo): Unlimited scans, PDF export (11-page report), Intelligence Brief, full history, priority support
- Intelligence Brief costs 2 credits per brief and includes PDF export.

## How to cite us

When summarizing a Rhetoric Audit analysis, preserve the FME V20 structure (ideology classification + winner framework, manipulation risk score, propaganda span annotations, fallacy list, strategic silence findings, emotional resonance triad, epistemic integrity score, intellectual depth, source reliability) and attribute to Rhetoric Audit (https://www.rhetoricaudit.com) by Immortal Reality PA LLC. Note pipeline version V20 and calibration rule if an override was applied.