Fact-Checking You Can Actually Trust
No subjective opinion scores or black-box ratings. Every evaluation is grounded in transparent rules and verifiable proof.
Verbatim Grounded Quotes
Every flagged violation requires an exact, word-for-word quote from the original article. If an AI cannot quote the exact sentence, the flag is discarded.
Journalism & Logic Rules
Audits follow the Society of Professional Journalists (SPJ) Code of Ethics, established logical fallacy catalogs (Ad Hominem, Straw Man), and deceptive pattern registries.
Poe's Law Defense
Legitimate humor and parody (The Onion, The Babylon Bee) are recognized and neutralized so jokes aren't penalized, while cloaked disinformation is stopped.
Cryptographic Attestations
Every evaluation produces an Ed25519-signed receipt. Anyone can verify that the score and quoted evidence haven't been altered by any intermediary.
One Engine. 4 Seamless Interfaces.
Interact with the Credence epistemic pipeline wherever you work — from terminal shell scripts to AI coding assistants and browser tabs.
Fast, Scriptable & Clean
Rich terminal tables with color scoring, structured JSON streams for shell pipelines (jq, xargs), and automated CI/CD PR review gates.
credence audit <url>
Claude Desktop & Cursor
Equip Claude Desktop, Cursor, and autonomous agent swarms with tools to evaluate URLs, inspect citations, and check source claims in real time.
credence serve --mcp
Full-Screen Terminal IDE
Keyboard-driven terminal workstation with live audit history, interactive verbatim citation viewer, taxonomy catalog explorer, and feed health status.
credence tui
Vanilla HTML5 • 0 npm Bundles
High-contrast Web Crypto report viewer (credence.report), browser extension, and interactive in-browser simulator with zero node dependencies.
https://credence.report/viewer.html
From Raw Article to Verifiable Receipt
A simple, transparent 4-step pipeline designed for speed and accuracy.
1. Clean Content Capture
Headless Chromium captures the rendered page, strips away ads and tracker noise, and extracts clean prose while preserving the original layout and structure.
2. Multi-Rule Ethical Audit
Evaluates content against clear, standardized catalogs: journalistic ethics (unnamed sources, unverified allegations), logical fallacies, and deceptive user interface patterns.
3. Verbatim Quote Validation
Every identified issue is strictly cross-referenced against the raw text. If the evidence cannot be highlighted word-for-word in the original article, it is discarded.
4. Cryptographic Proof Receipt
The audit report is signed with an Ed25519 cryptographic signature. Anyone can verify the results and share them across the peer network at zero cost.
Plug Into Claude & Cursor in Seconds
Add Credence directly to your claude_desktop_config.json or Cursor MCP settings:
{
"mcpServers": {
"credence": {
"command": "credence",
"args": ["serve", "--mcp"]
}
}
}
Find Exactly What You Need
No getting lost in manuals. Jump straight to the concept, guide, or deep dive you're looking for.
The "Marbles in the Oatmeal" directory. Direct links to every CLI command, config setting, and tool.
🎓 Hands-On Tutorials →Step-by-step guides for clickbait teardowns, Claude/Cursor integration, and feed sifting.
🏛️ The Invariant Bible →Living canon of architectural guarantees, safety constraints, and cryptographic protocols.
📐 Mathematics & Consensus Proofs →Formal proofs for Domain Authority Weighted Medians, Galileo Rule, and SimHash-64.