Rhetoric Audit

**Rhetoric Audit — PhD-Level Forensic Media Analysis in Your Browser**

Detect bias, propaganda, logical fallacies & strategic omissions — 15 diagnostic metrics in 2–5 seconds.

Rhetoric Audit is a production-verified forensic media analyzer that reads what an article says AND how it's engineered to manipulate you. Click the floating pill on any news article and receive a full forensic report instantly.

**What's new in v3.5.2**

Security & permissions release.

— Removed unused `scripting` permission (declared but never called — flagged in Chrome Web Store review)
— XSS hardening: all LLM-generated strings in the analysis panel are now HTML-escaped before rendering
— Permissions reduced to minimum necessary: `activeTab`, `storage`, `contextMenus` only

**What's new in v3.5.0**

This release upgrades the forensic engine to FME V20 — a unified single-call pipeline replacing the previous multi-stage system. Analysis is now faster, more consistent, and covers every article regardless of site structure.

New engine:
— FME V20 powered by GPT-4.1 Nano
— Universal article extraction via Mozilla Readability.js (now correctly analyses Reuters, EcoWatch, Hindustan Times, and other sites that use non-standard paragraph markup)
— Confidence score visible in the panel header at all times

Improved detection:
— Smarter content filter: category/hub pages (e.g. bbc.com/news/world) are no longer incorrectly scanned
— Subscription and account pages (e.g. pricing, userplan) blocked from triggering
— Reduced false negatives on legitimate news articles

Panel improvements:
— Evidence Quality score (replaces Evidence Validity Index) with calibrated anchors
— Factual Grounding removed (was not performing external verification — replaced by honest Evidence Quality signal)
— Heatmap now visible for all V20 scans
— Classification confidence shown as a header pill

Permissions:
This extension requests the minimum permissions necessary. v3.5.2 reduces permissions versus v3.5.0 — the `scripting` permission has been removed entirely. The only permissions requested are `activeTab`, `storage`, and `contextMenus`. The optional `clipboardWrite` permission is never used automatically — it only activates when you explicitly click the copy button.

**WHAT YOU GET:**

🎯 **Bias Spectrum Mapping** — Precision placement from Far Left to Far Right (not binary labels). Every score backed by reasoning.

💣 **Propaganda Index** — Quantifies emotional loading, manipulation tactics & ideological engineering (0–100%).

🚨 **24 Logical Fallacies** — Detects Ad Hominem, Strawman, False Dichotomy, Loaded Language, and 20 more — with exact quoted evidence.

🔇 **Strategic Silence** — Reveals what's deliberately omitted: missing counter-evidence, absent perspectives, ignored context.

😨 **Emotional Resonance Triad** — Separate Fear / Hope / Urgency scores to see what emotion the article targets.

📊 **Evidence Quality Index** — Verifiable data vs. anecdotal claims.

🧠 **Intellectual Depth (DOK)** — Webb's Depth of Knowledge framework (DOK-1 through DOK-4).

👁️ **Intent Transparency** — Is the author informing or obfuscating? (0–100).

✍️ **Source Reliability** — Author credibility based on citation patterns and epistemological humility.

🎭 **Narrative Archetype** — Hero Focus, Scapegoat Focus, or Victim Focus?

🏛️ **Philosophical Frame** — The underlying ideological lens of the piece.

**ACCURACY:**
- ✅ **100% band-accuracy** on 14-article production test suite (all 5 content strata)
- ✅ **Zero hallucinations** — all evidence quoted directly from article text
- ✅ **$0.003/article cost** — efficient ensemble model with fallback chain
- ✅ **Production-verified May 2, 2026** — 3 calibration patches deployed live

**HOW IT WORKS:**

1. **Install** the extension (free, no API key)
2. **Sign up** with email or Google (10 free analysis credits, no card required)
3. **Browse any news article** — a floating pill appears in the bottom-right corner
4. **Click the pill** → forensic panel opens in sidebar
5. **Export to PDF** for sharing or research

**FEATURES:**

✨ **Ambient Scanning** — Pill auto-appears on news articles (no manual trigger needed)
📄 **PDF Export** — Save full forensic reports with charts, claims ledger, and forensic synthesis
📊 **Dashboard Sync** — All analyses sync to your Rhetoric Audit dashboard
🔒 **Private** — Your analyses stay in your account; zero data sharing
⚡ **Fast** — 2–5 second analysis, even on long-form articles

**WHO IT'S FOR:**

🎓 **Researchers & Academics** — Pre-coded rhetorical analysis; save hours in literature reviews and media criticism
📰 **Journalists & Fact-Checkers** — Verify narrative framing, not just factual claims
📋 **Policy Analysts** — Track media framing on policy topics in real time
👁️ **Critical Readers** — Read ANY article with forensic clarity, not blind trust
🛡️ **Skeptics & Media Literacy Educators** — Teach students to detect bias & manipulation

**TECHNICAL:**

✅ Manifest V3 compliant
✅ Minimum permissions — `scripting` removed in v3.5.2; only `activeTab`, `storage`, `contextMenus` requested
✅ No inline scripts or eval()
✅ Content Security Policy enforced
✅ XSS-hardened panel — all dynamic content HTML-escaped before rendering
✅ CORS-validated API
✅ Rate-limited (anonymous: 10/day, free tier: 30/day)
✅ Supabase Row-Level Security — you only access your own analyses

**WHAT'S NEW IN v3.4.0:**

- Production API endpoint verified (www.rhetoricaudit.com)
- V19.1 forensic engine deployed (100% accuracy, 5 strata, MAE 2.5)
- Ambient floating pill on all news pages
- PDF export with full charts & claims ledger
- Dashboard sync for scan history
- Google OAuth + email sign-up
- Anon quota: 10 free analyses; sign up to unlock 30/day free tier

**PRIVACY & SECURITY:**

📋 [Privacy Policy](https://www.rhetoricaudit.com/privacy)
⚖️ [Terms of Service](https://www.rhetoricaudit.com/terms)
🌐 [OpenAPI Spec](https://www.rhetoricaudit.com/openapi.json)

---
Update: We have updated our privacy policy at https://www.rhetoricaudit.com/privacy to address all identified gaps:

Added explicit disclosure of the anonymous device identifier (ra_device_id) generated on install
Added a complete inventory of all chrome.storage.local keys and the data they contain
Clarified exactly which data fields are transmitted to each third-party (Supabase, Vercel) with per-party breakdowns
Updated the data transmission section to explicitly list article text, URL, page title, device ID, and user ID sent to our API

**Questions?** Visit https://www.rhetoricaudit.com or email hello@rhetoricaudit.com